DISCOVERY OF A STRONGLY PHASE-VARIABLE SPECTRAL FEATURE IN THE ISOLATED NEUTRON STAR RX J0720.4–3125

RX J0720.4–3125 belongs to a group of seven thermally emitting, radio-quiet, and nearby (≲500 pc) isolated neutron stars, originally discovered in ROSAT all-sky survey data, and often referred to as the X-ray dim isolated neutron stars (XDINSs). They are among the closest known neutron stars, and are characterized by X-ray luminosities ${L}_{{\rm{X}}}\approx {10}^{30-33}$ erg s⁻¹, long spin periods ($P\sim 3$–11 s), and inferred surface dipolar magnetic fields ${B}_{\mathrm{dip}}\approx {10}^{13}$ G, partially overlapping those of the magnetars (see van Kerkwijk & Kaplan 2007 and Turolla 2009 for reviews). XDINSs have estimated ages of a few 10⁵ years, derived from cooling curves and kinematics, while the characteristic ages are somewhat longer (∼10⁶ years). The X-ray spectra show thermal emission with inferred temperatures kT ≈ 50–100 eV and lack the non-thermal, power-law component often observed in other isolated neutron stars. Although a blackbody model provides an overall good description of their X-ray spectra, broad absorption features have been found in six of the XDINSs with XMM-Newton observations. Their origin is unclear: they can be produced either by proton cyclotron resonances/atomic transitions in a magnetized atmosphere, or by an inhomogeneous surface temperature distribution, as recently suggested by Viganò et al. (2014).

RX J0720.4–3125, the second brightest member of the class, was discovered by Haberl et al. (1997) as an isolated, pulsating neutron star with an 8.39 s spin period. The X-ray spectrum is best modeled by a blackbody (${kT}\sim 85$ eV) plus a broad (∼70 eV wide) absorption feature centered at ∼270 eV (Haberl et al. 2004). A further, narrow (∼5 eV wide) line was identified in XMM-Newton Reflection Grating Spectrometer data at ∼570 eV (Hohle et al. 2012a and references therein), possibly due to Oxygen circumstellar/interstellar absorption. Unlike other XDINSs, RX J0720.4–3125 exhibits long-term variations in its timing and spectral parameters. In the period 2001–2003 the total flux stayed almost constant, whereas the blackbody temperature increased from ∼84 to over 94 eV, and also the pulse profile, the blackbody radius, and the equivalent width of the absorption feature changed. In the following years this trend appeared to reverse, with a decrease of the surface temperature. Two main explanations for this behavior were proposed: either free precession, which predicts a cyclic pattern (Haberl et al. 2006), or a glitch that occurred around MJD = 52866 ± 73 days (van Kerkwijk et al. 2007). According to the most recent analyses (Hohle et al. 2009, 2012b), the interpretation in terms of a single sudden event around the proposed glitch epoch seems favored.

In this Letter we reanalyze all the archival XMM-Newton observations of RX J0720.4–3125, performing a detailed phase-resolved spectral analysis. In Section 2 we report on the data analysis, and present the discovery of a second, new, phase-dependent absorption feature in the X-ray spectrum, the possible origin of which is discussed in Section 3.

RX J0720.4–3125 was observed 20 times by the XMM-Newton satellite using the European Photon Imaging Camera (EPIC). We used here only data obtained with the EPIC-pn camera (Strüder et al. 2001) because less affected by pile-up (see Table 1). We processed the data using the Science Analysis Software (sas, version 13.5.0), adopting the most recent calibration files available, and we removed any particle flare via good-time-intervals. All arrival times were referred to the Solar System barycenter (source coordinates ${\rm{R}}.{\rm{A}}.={07}^{{\rm{h}}}{20}^{{\rm{m}}}24\buildrel{\rm{s}}\over{.} 961$, $\mathrm{decl}.=-31^\circ 25\prime 50\buildrel{\prime\prime}\over{.} 21$, J2000.0; Kaplan et al. 2003), and a rotational phase was assigned to the source counts of all the observations using the timing solution of Hohle et al. (2012b). For the source distance we assumed D = 286 pc (Tetzlaff et al. 2011). All errors are reported at the 90% confidence level.

Spectral analysis was performed using the XSPEC analysis package (version 12.8.2; Arnaud 1996), using a minimum of 100 counts per bin for each spectrum, the ${\chi }^{2}$ statistics and a maximum oversampling of the spectral energy resolution of a factor of 3. The fits were restricted to energies between 0.3 and 1.2 keV to exclude the broad absorption line at ∼270 eV so as to reduce the number of degrees of freedom (dof). We extracted the source spectra from a circular region of radius $30\prime\prime$ centered on the source point-spread function and the background counts from a circle of the same size far from the source and on the same CCD. For small window and full frame observations the pile-up level, estimated by means of the sas epatplot tool, was <1.5% and 4%, respectively (see Table 1). To mitigate the pile-up we restricted our spectral analysis to photons having flag = 0 and pattern = 0.

We started our analysis from the 2003 May 2 observation, this being the longest observation, and least affected by pile-up (the pn was operating in Small Window mode). The phase-averaged spectrum is well fitted by an absorbed¹⁰ blackbody (reduced chi-square ${\chi }_{\nu }^{2}$ = 1.17 for 130 dof) yielding the following parameters: column density ${N}_{{\rm{H}}}$ $\;=\;1.9(3)\times {10}^{20}$ cm⁻², blackbody temperature ${{kT}}_{\mathrm{BB}}$ = 82.1 ± 0.6 eV and blackbody radius ${R}_{\mathrm{BB}}$ = 6.3 ± 0.6 km.

To study in detail the spectral variability with phase, we produced a phase-energy image by binning the EPIC-pn source counts into 100 rotational phase channels and 25 eV wide energy channels, and then normalizing to the phase-averaged energy spectrum. The image (Figure 1) shows a feature in the phase interval 0.1–0.3, which gives a strong hint for the presence of an absorption line at ∼750 eV in the corresponding spectrum. Above ∼1 keV the source counts become background dominated.

Zoom In Zoom Out Reset image size

To better investigate the significance of the feature appearing in Figure 1, we performed a phase-resolved spectral analysis for the 2003 May 2 observation dividing the rotational cycle in five equal phase bins. For the spectrum relative to phase 0.1–0.3 the addition of a Gaussian line in absorption to the blackbody continuum leads to a significant improvement in the shape of residuals, while a simple blackbody model gives an acceptable fit for all the other phase bins. Both the gabs and Gauss models result in ${\chi }_{\nu }^{2}$ = 0.88 for 89 dof, corresponding to an F-test probability of 3.5 × 10⁻⁷. We multiplied this value by 5, considering the five phase-intervals as trials. In this way we obtained a significance of $\sim 5\sigma$. The values for line energy and width, obtained from the two models, are compatible within the errors. The best-fit tbabs*(bbodyrad+Gauss) model gave the following parameters: ${{kT}}_{\mathrm{BB}}$ = 83.0 ± 0.1 eV, ${R}_{\mathrm{BB}}$ = 6.3 ± 0.3 km, line energy E_line = ${745}_{-27}^{+17}$ eV, width σ = ${42}_{-33}^{+51}$ eV and normalization of $-{9.2}_{-9.2}^{+3.5}\times {10}^{-5}$ (the column density was frozen at the value obtained for the phase-averaged spectrum). The equivalent width of the feature is ${28}_{-11}^{+9}$ eV. Table 2 summarizes the results of the phase-resolved spectroscopy for the 2003 May 2nd observation using the tbabs*(bbodyrad+Gauss) model. Figure 2 shows the phase-resolved spectrum relative to phase 0.1–0.3 (black) in comparison with that at phase 0.5–0.7 (red): the absorption feature clearly shows up in the residuals.

Zoom In Zoom Out Reset image size

Table 2.  Pulse Phase Spectroscopy For 2003 May 2 Observation

Parametera	0.1–0.3	0.3–0.5	0.5–0.7	0.7–0.9	0.9–1.1
BB
${{kT}}_{\mathrm{BB}}$ (eV)	81.1(6)	81.9(6)	82.2(6)	82.6(6)	82.5(9)
${R}_{\mathrm{BB}}$ (km)	${6.4}_{-0.5}^{+0.6}$	${6.0}_{-0.5}^{+0.6}$	${6.1}_{-0.5}^{+0.6}$	${6.4}_{-0.5}^{+0.6}$	${6.4}_{-0.6}^{+0.7}$
Fluxb	0.86(1)	0.82(1)	0.86(1)	0.95(1)	0.95(1)
Unabs. Fluxb	1.07	1.01	1.06	1.17	1.17
NHPd	$7.2\times {10}^{-2}$	$5.9\times {10}^{-1}$	$6.9\times {10}^{-1}$	$4.2\times {10}^{-2}$	$2.7\times {10}^{-1}$
${\chi }_{\nu }^{2}$	1.22	0.96	0.92	1.26	1.09
dof	92	90	92	97	75
BB+GAUSS
${{kT}}_{\mathrm{BB}}$ (eV)	${83.0}_{-0.9}^{+1.5}$	82.4(9)	${82.2}_{-0.6}^{+0.7}$	82.7(8)	${82.4}_{-0.9}^{+1.0}$
${R}_{\mathrm{BB}}$ (km)	6.0(6)	${6.0}_{-0.5}^{+0.6}$	${6.1}_{-0.5}^{+0.6}$	${6.3}_{-0.5}^{+0.6}$	6.4(6)
${E}_{\mathrm{line}}$ c (eV)	${745}_{-27}^{+17}$	745	745	745	745
σ c(eV)	${41.7}_{-33.8}^{+51.3}$	41.7	41.7	41.7	41.7
$| \mathrm{Norm}|$	${9.2}_{-9.2}^{+3.5}\times {10}^{-5}$	≤2.7 × 10−5	≤1.7 × 10−5	≤2.9 × 10−5	≤2.4 × 10−5
Eq. Width (eV)	${28}_{-11}^{+9}$	${6}_{-5}^{+9}$	$\leqslant 8$	$\leqslant 11$	$\leqslant 13$
F-test (10−3)	$3.5\times {10}^{-7}$	232	1000	750	>1000
NHPd	$7.9\times {10}^{-1}$	$6.0\times {10}^{-1}$	$6.6\times {10}^{-1}$	$3.6\times {10}^{-2}$	$2.5\times {10}^{-1}$
${\chi }_{\nu }^{2}$	0.88	0.95	0.93	1.27	1.11
dof	89	89	91	96	74

Notes.

^a ${N}_{{\rm{H}}}$ was frozen at the value obtained for the phase averaged spectra: ${N}_{{\rm{H}}}$= $1.9\times {10}^{20}$ cm⁻² . ^bFluxes are calculated in the 0.3–1.2 keV energy range and in units of (10⁻¹¹ erg s⁻¹ cm⁻² ). ^cLine energy and width were frozen at the value obtained for the phase interval 0.1–0.3: E_line = 745 eV and σ = 42 eV. ^dNHP is the Null Hypothesis Probability.

Download table as:  ASCIITypeset image

The same phase-averaged and phase-resolved analysis was then repeated for all the remaining XMM-Newton observations in search for the presence of the 745 eV absorption line. Since observations in Full Frame mode were affected by pile-up, their spectra were extracted from an annular region extending from 15'' to 30'' from the nominal source position, this time including also events with $\mathrm{PATTERN}\;\leqslant$ 4. The resulting values of the line equivalent width and the F-test probability as a function of time are shown in Figure 3 (top panel), and clearly indicate that the presence of the feature is unconstrained in most of the observations, possibly because of their lower statistics.

Zoom In Zoom Out Reset image size

We then performed a simultaneous fit of the spectra in the 0.1–0.3 phase bin for all the observations in Full Frame mode with the Thin filter (18 spectra in total; see Table 1). We constrained ${N}_{{\rm{H}}}$ and the line energy to be the same (to minimize the free parameters), and the width of the feature was frozen at the value obtained for the 2003 May observation. The phase-dependent feature is detected at an energy of 785(13) eV for the tbabs∗(bbodyrad+Gauss) model (${\chi }_{\nu }^{2}$ = 0.96 for 556 dof); normalizations and equivalent widths are consistent within the errors in all the spectra. The tbabs∗bbodyrad model gives ${\chi }_{\nu }^{2}$ = 1.03 for 575 dof. The F-test probability for the inclusion of the Gaussian absorption line is $4.2\times {10}^{-6}$, which corresponds to a significance level of $\sim 4.5\sigma$ (taking into account the number of trials).

To further test the presence of this feature in data taken before and after 2003 May 2, we also combined all the spectra except that of 2003 May 2. We used the SAS epicspeccombine tool to merge the spectra and the response matrices. For the combined spectrum in the phase range 0.1–0.3, the best-fit with a blackbody model yields ${\chi }_{\nu }^{2}$ = 2.04 for 123 dof; by including a Gaussian absorption line, ${\chi }_{\nu }^{2}$ decreases to 1.48 for 121 dof. The parameters for the phase-dependent line are: E_line = 795 ${}_{-14}^{+15}$ eV, normalization of (−8.3 ${}_{-1.7}^{+1.7}$) × 10⁻⁵and equivalent width of ${27}_{-5}^{+6}$ eV for the Gauss model (the line width was fixed to the value obtained from the 2003 May 2 observation). The F-test probability between these two models is $1.3\times {10}^{-9}$ and the significance is equal to 6σ (again considering 5 trials).

As a further check, we built a combined spectrum comprising all observations. This spectrum, created by merging all the spectra extracted from a 15'' to 30'' annular region with $\mathrm{PATTERN}\;\leqslant$ 4, has an exposure time of 406.8 ks and a much improved statistics. The 0.1–0.3 phase-resolved spectrum shows an absorption feature at energy ${787}_{-14}^{+15}$ eV with normalization of $-{7.9}_{-1.7}^{+1.6}\times {10}^{-5}$ and equivalent width of ${26}_{-5}^{+4}$ eV for the Gauss model; also in this case we froze the line width at the value estimated for the longest observation. The F-test probability for the inclusion of the absorption feature is now $4.7\times {10}^{-9}$ and, taking into consideration the number of trials, the significance level is again $\sim 6\sigma$. As shown in Figure 2, panel (d), the spectral feature is clearly detectable in the fit residuals of the merged observations, and the line significance increases with respect to the 2003 May observation alone.

In the spectra relative to the other phase intervals the line equivalent width is always $\lt 13$ eV, and the F-test probability is never $\lt 3\times {10}^{-3}$ (see Figure 3, lower panel). We stress that the response of the pn camera has slightly changed (see, e.g., Sartore et al. 2012) over the 12 years spanned by the observations, hence mixing spectra obtained with different settings over such a long period certainly introduce large systematic errors which are difficult to quantify. However, a spectral feature present only in a given phase interval cannot result from systematic effects, which are definitely independent on the rotational phase. While the line is mostly evident in the 2003 May 2 observation, which we used to derive the best estimate of the line parameters, the considerations presented above strongly suggest that the feature was present also in the other observations, with properties which appear relatively stable over the 12 year period covered by XMM-Newton data.

A careful re-analysis of all available XMM-Newton observations of the isolated neutron star RX J0720.4–3125 revealed the presence of a phase-dependent absorption feature at ∼750 eV, which is mostly evident during the decline of the pulse profile (phases 0.1–0.3). The feature was first significantly detected (at $\sim 5\sigma$ confidence level) in the 2003 May 2 XMM-Newton observation, which was the longest one and was less affected by pile-up. Phase-dependent spectral analysis of the remaining datasets, which cover the period 2000 May–2012 September, indicates that the feature is likely present at all epochs and its properties are consistent with being constant over the timespan covered by available observations (∼12 years). This suggests that the feature may be long-lasting and not associated with the timing anomalies (possibly due to a glitch) that occurred in 2003 (van Kerkwijk et al. 2007; Hohle et al. 2012b).

Interestingly, an absorption feature with somewhat similar properties has been discovered in the spectrum of the "low-field" magnetar SGR 0418+5729 (${B}_{\mathrm{dip}}\sim 6\times {10}^{12}$ G; Tiengo et al. 2013). In both cases the feature is present only in a quite narrow phase interval ($\sim 1/5$ of the phase cycle), but in SGR 0418+5729 the line energy is higher ($\gtrsim 2$ keV) and strongly variable in phase (by a factor ∼5 in one-tenth of the phase cycle). No clear variation of the line energy with phase is seen in RX J0720.4–3125, although this might simply reflect the fact that the source counts become background dominated at energies just above that of the feature. Tiengo et al. (2013) interpreted the feature in SGR 0418+5729 as due to proton cyclotron resonant scattering in a baryon-loaded magnetic loop close to the star surface. The B-field variation along the loop explains the strong dependence of the line energy with phase, since photons emitted from the surface intersect the loop at different positions.

The similarities between the two features are suggestive that the same mechanism may be at work in RX J0720.4–3125. If the feature is a (broadened) proton cyclotron line, its energy is ${E}_{\mathrm{line}}=0.63({B}_{\mathrm{loop}}/{10}^{14}\;{\rm{G}})/(1+z)$ keV, where 1/(1 + z) ∼ 0.7 is the gravitational redshift at the star surface. The implied magnetic field in the loop is ${B}_{\mathrm{loop}}\sim 1.8\times {10}^{14}$ G, a factor of ∼7 larger than the the equatorial value of the dipole magnetic field at the surface, ${B}_{\mathrm{dip}}\sim 2.5\times {10}^{13}$ G. A preliminary calculation using the same geometrical model as in Tiengo et al. (2013) shows that the energy of the line and its appearance only for ∼20% of the star period can indeed be reproduced. It should be noted, however, that the exact geometry of the magnetic structure is likely different in the two sources. In the case of SGR 0418+5729, in fact, X-ray photons come from a quite small, hot spot with size ∼1 km, while the emitting region of RX J0720.4–3125, albeit not uniform, is larger (${R}_{\mathrm{BB}}\sim 6$ km).

A further possibility is that the line results from atomic absorption in the neutron star (magnetized) atmosphere. The strong dependence of the feature on the star rotational phase may be due to local inhomogeneities in the star surface magnetic field (as in the cyclotron scenario above) and/or to a clumpy structure of the atmosphere itself. The possibility that primary photons come from a small hot spot appears, on the other hand, unlikely, given the relatively large radiation radius and the broad shape of the pulse profile. If the feature arises from bound-bound (or bound-free) transitions, the atmosphere composition is most probably He or mid-Z elements (see, e.g., Pavlov & Bezchastnov 2005; Mori & HO 2007). The ionization energy for H is, in fact, below ∼0.7 keV for fields $\lesssim {10}^{14}\ {\rm{G}}$ (above this value the feature may be washed out by vacuum resonance mode conversion; e.g., van Kerkwijk & Kaplan 2007 and references therein).

The presence of proton cyclotron lines produced by resonant scattering/absorption in confined, high-B structures close to the star surface, where the magnetic field is a factor ≳10 higher than the dipole, would be supportive of a picture in which the magnetic field of (highly magnetized) neutron stars is complex, with substantial deviations from a pure dipole on the small scales. In recent years, a better (although far from conclusive) understanding of how stellar magnetic fields are generated and amplified indicates that strong, non-dipolar field components are ubiquitous, from massive stars (Braithwaite & Spruit 2006) to proto-neutron stars (Obergaulinger et al. 2014). It is now well established that internal toroidal field components, multipolar surface structures, as well as very localized B-field bundles must be present in many neutron stars, especially in highly magnetized ones. The recent discovery of sources showing strong magnetic-powered flares and outbursts, but with low dipolar magnetic fields, $\lesssim {10}^{13}$ G, further strengthens this idea (Rea et al. 2010, 2012, 2014; Turolla & Esposito 2013).

Finally, we stress that the absorption feature reported here for the first time is unlikely related to the broad absorption feature at 270 eV (Haberl et al. 2004, 2006). Even if the two may have the same physical origin, i.e., a proton cyclotron resonance, the latter implies a lower magnetic field, $\sim 5\times {10}^{13}$ G, comparable to the spin-down estimate, ${B}_{\mathrm{dip}}\sim 2.4\times {10}^{13}$ G, indicating that in this hypothesis it is related to the large-scale field. Actually broad absorption features have been detected in all the XDINSs, with the exception of RX J1856.5–3754 (e.g., Turolla 2009), but the strong phase variability we observe for the 0.75 keV line of RX J0720.4–3125 sets it apart.

This is the first time evidence of a complex magnetic field structure is observed in an XDINS, although this result is not unexpected, if, as suggested by many investigations, XDINSs are aged magnetars (Viganò et al. 2013; see also Turolla 2009).

A.B., N.R., and F.C.Z are supported by an NWO Vidi Grant (PI: Rea) and by the European COST Action MP1304 (NewCOMPSTAR). N.R. acknowledges support by grants AYA2012-39303 and SGR2014-1073. The work of R.T. and A.T. is partially supported by an INAF PRIN grant.