A NEW γ-RAY LOUD, ECLIPSING LOW-MASS X-RAY BINARY

An X-ray source in this region was first detected in the 20–60 keV band by INTEGRAL as IGR J04288–6702 (Grebenev et al. 2013). Follow-up 0.3–10 keV Swift/XRT observations resolved this source into two separate sources visible in softer X-rays (Figure 1). A hard (14–195 keV) detection in this region was also reported by the Swift/BAT 70-month survey (Baumgartner et al. 2013). Here the source is noted as a close "double" source, suggesting the possibility that both Swift/XRT sources may have been detected by Swift/BAT. The Swift/BAT 14–195 keV flux is $1.5\times {10}^{-11}$ erg s⁻¹ cm⁻², and the best-fit power law has a photon index of ${\rm{\Gamma }}=0.95$. However, this flux value should be interpreted with caution due to the blended nature of the source.

Zoom In Zoom Out Reset image size

The brighter of the Swift/XRT sources (1SXPS J042947.1–670320; mean count rate $(5.6\pm 0.3)\times {10}^{-2}$ cts sec⁻¹) has an absorbed power law with ${\rm{\Gamma }}=1.8\pm 0.2$, N_H consistent with the Galactic value of $5\times {10}^{20}$ cm⁻², and an unabsorbed flux of $2.3\pm 0.2\times {10}^{-12}$ erg s⁻¹ cm⁻² (Evans et al. 2014). To study the fainter source (1SXPS J042749.2–670434; mean count rate $(1.2\pm 0.2)\times {10}^{-2}$ cts sec⁻¹) in more detail, we analyzed 10 ks of Swift/XRT observations taken in 2010 October (P.I. Markwardt). Using the task xrtgrblc of HEAsoft (v6.19), we extracted the spectrum of the source, which can be described by a hard absorbed power law with parameters of ${N}_{H}\,={2}_{-2}^{+4}\,\times {10}^{22}$ cm⁻², ${{\rm{\Gamma }}}_{\mathrm{sw}}={1.0}_{-1.6}^{+2.1}$, and ${F}_{0.3\mbox{--}10\mathrm{keV}}\,={2.0}_{-0.5}^{+15.9}\,\times {10}^{-12}$ erg s⁻¹ cm⁻² (${\chi }_{\nu }^{2}=2.50/1;$ all uncertainties are in 90% confidence). Because only about 80 useful counts were collected, the XRT light curve is too noisy to show any significant variability. We note that an additional Swift/XRT observation was taken simultaneously with the NuSTAR data (Section 2.1.4), but with only five net counts, these data were not useful for analysis, though we do analyze the associated Swift/UVOT photometry in Section 2.2.3.

A γ-ray source in this region is present in the in the Fermi/LAT 3FGL catalog (Acero et al. 2015), listed as 3FGL J0427.9–6704, with J2000 R.A. and decl. of (04:27:55.7, –67:04:40). The positional error ellipse is nearly circular, with a 95% radius of $\sim 4\buildrel{\,\prime}\over{.} 2$. The catalog flux of the source is (9.4 ± 0.8) $\times \,{10}^{-12}$ erg s⁻¹ cm⁻² (0.1–100 GeV). The γ-ray spectrum is well-fit by a power law with an index of ${\rm{\Gamma }}=2.46\pm 0.07$. While the detection $\gt 10\,\mathrm{GeV}$ is marginal, there is no evidence of curvature, and the flux in this band is consistent with the power law at lower energies. We have also analyzed Pass 8 observations of this region with very similar spectral results and an improved localization (Section 3.3.3).

As Figure 1 shows, the 20–60 keV INTEGRAL position is close to halfway between the two Swift/XRT sources, suggesting that they have comparable flux at these energies (and unlike in the soft 0.3–10 keV XRT band, where the count rates are different by a factor of approximately five). The Swift/BAT 14–195 keV position is yet closer to the fainter XRT source. Comparing the location of the Fermi/LAT γ-ray source to the two Swift/XRT sources: the fainter source is near the center of the 3FGL γ-ray error ellipse, while the brighter XRT source sits nearly $\sim 11^{\prime}$ away (Figure 1), far outside the Fermi 3FGL error ellipse. There are no other significant X-ray sources within the γ-ray error ellipse.

All of the high-energy observations are consistent with an interpretation in which there are two X-ray sources in this region with differing spectra, such that the fainter Swift/XRT source is also harder. It gradually overtakes the flux of the brighter Swift/XRT source at higher energies, and in γ-rays is the only source detected. In this scenario, all of the Fermi/LAT flux and much of the blended Swift/BAT flux comes from this source, which we call 3FGL J0427.9–6704 henceforth.

The high-resolution Swift/XRT data allow the identification of optical counterparts to the X-ray sources. To within 2'', the Swift/XRT position of 3FGL J0427.9–6704 matches a moderately bright point source (J2000 R.A. and decl. of 04:27:49.61, –67:04:35.0) with $R\sim 17.1$ (Monet et al. 2003). As we will show in Section 3, this source shows all the optical properties of an accreting Galactic compact binary and is thus nearly certain to be associated with the X-ray and γ-ray emission discussed here. The brighter, softer Swift/XRT source is located at the center of a distant galaxy and is therefore a likely active galactic nucleus.

For follow-up observations of the Swift/XRT source associated with 3FGL J0427.9–6704, on 2016 May 19, we obtained 84 ks of observations with NuSTAR (Harrison et al. 2013), with 3FGL J0427.9–6704 placed at the nominal aimpoint. Of the 84 ks, 60 ks were on-source. 3FGL J0427.9–6704 is clearly detected with a net count rate of $(4.89\pm 0.08)\times {10}^{-2}$ cts s⁻¹ per detector (about 6800 source counts in total). These data are analyzed in Section 3.3.

While the location of the source is west of the main body of the LMC, it is covered in the ongoing OGLE-IV survey, which has added fields to monitor the bridge area between the Magellanic Clouds (Udalski et al. 2015). We converted the OGLE instrumental magnitudes i and v into standard I and V filters using the equations $I=i-0.338-0.0047(v-i)$ and $V=v-0.066\mbox{--}0.072(v-i)$, with $(v-i)$ assumed from a smoothed version of the light curve for epochs with only one filter available. We also converted the OGLE UTC HJD values into Barycentric Julian Date (BJD) on the Barycentric Dynamical Time (TDB) system (Eastman et al. 2010). The OGLE photometry comprise 368 epochs in I and 28 epochs in V, spanning 2010 March 07 to 2015 November 29.

From 2016 January 16 to May 1 (UT), we obtained optical BVI and near-IR H photometry of the source on most clear nights using ANDICAM on the SMARTS 1.3 m telescope at CTIO. On a few nights multiple sets of observations were taken; the total number of epochs was 62–64 per band, depending on the band. The exposure time in each BVI was 300 s; the on-source exposure time in H varied but was typically 500–600 s. The data were reduced as described in Walter et al. (2012). We performed differential aperture photometry of the source with respect to 10 (BVI) or 4 (H) bright stars in the field, with absolute calibration via Landolt (1992) standards on clear nights (BVI) or with respect to 2MASS (H). All the OGLE and SMARTS photometry can be found in Table 1.

We obtained UV photometry from Swift/UVOT observations of the field. In the 2010 and 2016 Swift observations, the object is detected in all of uvw1, uvm1, uvw2, and u. For the near-IR, we retrieved 2MASS JK and WISE W1/W2 measurements from the 2MASS 6x (Cutri et al. 2012) and AllWISE (Cutri et al. 2014) catalogs, respectively. To facilitate modeling of the mean spectral energy distribution of the system, we list the mean out-of-eclipse magnitudes for all bands in Table 2. We emphasize that any such modeling should be undertaken with the knowledge that the UV, optical, and IR emission is variable and that our UV and IR data (excepting H) only represent one to a few epochs.

Table 2.  Mean Panchromatic Photometry^a

Filter	AB mag
uvw2	19.84 ± 0.17
uvm2	19.77 ± 0.03
uvw1	19.44 ± 0.12
u	18.51 ± 0.22
B	18.01 ± 0.03
V	17.72 ± 0.03
I	17.27 ± 0.01
J	17.05 ± 0.05
H	17.13 ± 0.02
K	17.28 ± 0.09
W1	18.25 ± 0.03
W2	18.75 ± 0.06

Note.

^aFor bands with multiple observations, the mean magnitudes out of eclipse (taken as $\phi \pm 0.05$ of each eclipse) and standard errors of the mean are given. These magnitudes are not corrected for extinction.

Download table as:  ASCIITypeset image

Given that the eclipsing nature of the system implies a close to edge-on orientation, velocity measurements could, in principle, allow a good constraint on the mass of the compact object primary in the binary. Therefore, we obtained a large number of spectra with the medium-resolution 1200 l mm⁻¹ grating to measure precise radial velocities for the secondary. As we explain in detail below, the results were unexpected.

Figure 4 shows a trailed spectrograph of medium-resolution spectra over the wavelength range of 4790–5300 Å in bins of 0.02 in phase and smoothed by a 5 pixel boxcar. The strong double-peaked Hβ emission and deep He i absorption lines at 4922 and 5016 Å are the most obvious features (absorption associated with Mgb around 5170 Å also appears to be present). Qualitatively, these absorption lines match the Hβ emission in both velocity amplitude and phase. However, there is a forest of weaker absorption lines that were not obvious in the examination of the low-resolution spectra. These weaker lines have a much larger radial velocity amplitude and are offset in phase from the strong absorption and Hβ emission lines. These findings lead to the unexpected conclusion that the strong absorption lines are associated with the disk rather than the secondary, whose motion is instead traced by the weaker absorption lines.

Zoom In Zoom Out Reset image size

To put these conclusions on a more quantitive basis, we measured the strong absorption line radial velocities using the He i lines at 4922 and 5016 Å, which are the strong lines least affected by emission or confusion with other lines. The velocities were determined through cross-correlation with a spectrum of the B3V star HD 210121 taken with the same instrumental setup. To measure the weaker lines, we selected the cleanest, most isolated regions containing these lines, which we judged to be 4950–4970 Å and 5195–5285 Å. Of these lines, several can be clearly attributed to Fe i and Cr i, suggesting a mid-to-late K spectral type of the secondary. We determined the radial velocity from these weak lines through cross-correlation with a spectrum of the K7V star HD118100.

Our spectral range was chosen to contain Hβ and nearly all spectra show double-peaked Hβ emission likely formed in the outer regions of the accretion disk. To measure the orbital motion of this disk, we performed a three-component fit to the spectrum in the region of Hβ, fitting the background and two Gaussian line components. In general, the approaching (blue) component of the line is somewhat weaker and slightly asymmetric and hence we focus on the receding (red) line for comparison with the absorption line radial velocities.

Figure 5 shows the strong and weak absorption line and red emission-line radial velocities plotted as a function of orbital phase, where we have anticipated the results of Section 3.3 and assumed that the I-band eclipse corresponds to superior conjunction of the compact object, which we set as ϕ = 0.75 (in this phase convention the ascending node of the compact object is ϕ = 0.5). To better compare the emission-line to strong absorption line velocities, we have added an offset to the emission-line velocities, as only the red component of the Hβ emission is plotted. Outside of eclipse, this red component is consistent with a circular Keplerian orbit peaking at $\phi \sim 0.5$, as expected if the disk surrounds the compact object that is eclipsed at ϕ = 0.75.

Zoom In Zoom Out Reset image size

Remarkably, the strong absorption line radial velocities also peak at $\phi \sim 0.5$, matching the emission. This is wholly inconsistent with the natural association of absorption lines with the secondary, whose lines would be expected to peak at ϕ = 0. The interpretation most consistent with these data is that the strong absorption lines are associated with the disk, not the secondary, matching our initial finding on the basis of the trailed spectrogram in Figure 4. We note that the absorption line radial velocity curve does not have a simple Keplerian shape, with a minimum closer to $\phi \sim 0.25$ than 0 as expected in our proposed scenario, and a quick rise to the maximum at $\phi \sim 0.5$.

The strong absorption lines appear resolved, with a typical full-width at half-maximum of ∼300–400 km ${{\rm{s}}}^{-1}$, much narrower than the emission lines (Sections 3.2 and 3.4). This is consistent with the formation of these absorption lines in a restricted region of the cool outer disk along the line of the sight to the hot inner disk. However, the detailed formation of these unusual absorption lines (apparent at nearly all epochs) may be complex and we do not model them further in this paper. To our knowledge, optical absorption lines have never previously been observed in a low-mass X-ray binary accreting in the low state.

In contrast to the rather noisy radial velocity curves of the emission and strong absorption lines, the velocities of the weak lines show a clean circular Keplerian curve that peaks at ϕ = 0, exactly out of phase with the other curves. This confirms that these lines arise in the faint secondary and trace its motion around the binary's center of mass, and indeed that this system is the equivalent of an eclipsing double-lined spectroscopic binary.

To characterize these orbits, we fit circular Keplerian models to the radial velocities with the period and time of conjunction fixed to the values determined from the photometry. Focusing first on the red emission-line radial velocities: as the rms at each phase is evidently much larger than expected on the basis of the formal uncertainties, we first did an unweighted fit, finding a semi-amplitude ${K}_{\mathrm{em}}=103\pm 8$ km s⁻¹ and a systematic velocity (for this line) of ${v}_{\mathrm{red}}=596\pm 6$ km s⁻¹. This fit had an rms of 52 km s⁻¹ (compared to a mean formal uncertainty of 21 km s⁻¹). A weighted fit yields ${K}_{\mathrm{em}}=106\pm 5$ km s⁻¹ and ${v}_{\mathrm{red}}\,=592\pm 4$ km s⁻¹, with an rms of 54 km s⁻¹. Because the relative uncertainties probably do capture information about the fitting, it may be that the best estimate lies between these listed values, but for the remainder of the paper we assume the former, unweighted values.

As noted above, the absorption line radial velocity curve is not well-fit by a Keplerian curve in the range of $\phi \sim 0.2$–0.4. Nonetheless, we performed such a fit as a check on the emission-line model. The fit rms is higher as expected (70 km s⁻¹), however, the best fit is ${K}_{\mathrm{em}}=99\pm 11$ km s⁻¹, entirely consistent with that from the emission-line fit.

For the weak absorption lines that trace the secondary, we find the secondary semi-amplitude ${K}_{2}=293\pm 4$ km s⁻¹ and ${v}_{\mathrm{sys}}=79\pm 3$ km s⁻¹. The rms of this fit is 14 km s⁻¹, much smaller than that for disk velocities, but still somewhat higher than expected on the basis of the individual uncertainties, probably because of a varying disk contribution to the regions used for cross-correlation. One potential concern is that K₂ is biased low from the observed center-of-mass motion due to irradiation of the secondary. This can appear as a false eccentricity in the velocity curve (e.g., Davey & Smith 1992). To test whether this is occurring, we tried allowing a non-circular fit, but found that this did not improve the fit. As a second test, we fit the radial velocity curves of the warm ($\phi =0$–0.5) and cold ($\phi =0.5$–1) sides of the secondary separately, finding K₂ values consistent to within 1%. In addition, the strength of the weak lines does not appear to appreciably vary over the orbit. We conclude that there is no evidence that K₂ is significantly affected by irradiation.

We analyzed the 84 ks of NuSTAR data using the nuproducts task of HEAsoft with a 60'' radius circular extraction region and a source-free 60'' background region, extracting barycentric-corrected light curves and spectra over the energy range of 3–79 keV using both focal plane modules A and B (FPMA/B). After spectral binning, the NuSTAR data with a minimum of 20 counts per bin, we fit them with an absorbed power law (i.e., const∗phabs∗cflux∗pow in XSPEC, using a multiplicative factor const for the cross-calibration between FPMA and FPMB: ${{\rm{C}}}_{\mathrm{FPMB}}={F}_{\mathrm{FPMB}}/{F}_{\mathrm{FPMA}}$). All fits used the abundance scale of Anders & Grevesse (1989). The best-fit parameters are ${{\rm{C}}}_{\mathrm{FPMB}}=1.05\pm 0.05$, N_H = ${5.8}_{-1.7}^{+1.8}\times {10}^{22}$ cm⁻², ${\rm{\Gamma }}={1.68}_{-0.08}^{+0.09}$, and ${F}_{3\mbox{--}79\mathrm{keV}}\,=(6.7\pm 0.4)\times {10}^{-12}$ erg s⁻¹ cm⁻² (${\chi }_{\nu }^{2}=254.0/280$). Both the photon index and the column density are consistent with that of the Swift/XRT spectrum taken in 2010 October, though with much smaller uncertainties. We note that this best-fit flux includes the X-ray eclipse (Section 3.3.2) and so the best estimate of the uneclipsed flux is 6% higher: ${F}_{3\mbox{--}79\mathrm{keV}}=(7.1\pm 0.4)\times {10}^{-12}$ erg s⁻¹ cm⁻². Given the low foreground extinction, the high N_H is likely due to material associated with the disk that produces photoelectric absorption due to the high inclination of the system.

As a check on the consistency between the NuSTAR and Swift data, we also fit the 2010 Swift and 2016 NuSTAR X-ray spectra together with an additional cross-calibration factor ${C}_{\mathrm{XRT}}={F}_{\mathrm{XRT}}/{F}_{\mathrm{FPMA}}$. The results are ${C}_{\mathrm{FPMB}}=1.06\pm 0.05$, ${C}_{\mathrm{XRT}}=0.9\pm 0.2$, N_H = ${4.4}_{-1.2}^{+1.6}\times {10}^{22}$ cm⁻², ${\rm{\Gamma }}={1.63}_{-0.07}^{+0.08}$, and ${F}_{3\mbox{--}79\mathrm{keV}}=(6.8\pm 0.4)\times {10}^{-12}$ erg s⁻¹ cm⁻² (${\chi }_{\nu }^{2}\,=260.3/283$). All of the best-fit parameters still remain within the original NuSTAR-only uncertainties. The Swift–NuSTAR cross-calibration factor is ∼0.9, consistent with no substantial X-ray flux change between the epochs. Nonetheless, given the possibility for modest variations in the system and the relatively low net counts in the Swift data, we only fit the NuSTAR data in the following spectral analysis.

Given the high column density (N_H $\,\gt \,{10}^{22}$ cm⁻²) it is difficult to constrain the presence of an additional minor thermal component in the spectrum. Formally, the addition of a blackbody component slightly improves the fit: a component with ${T}_{\mathrm{bb}}={5.6}_{-1.7}^{+1.8}$ keV and ${F}_{3\mbox{--}79\mathrm{keV}}={1.0}_{-0.8}^{+0.9}\times {10}^{-12}$ erg s⁻¹ cm⁻² results in ${\rm{\Delta }}{\chi }^{2}=-4.3$ with two extra degrees of freedom. However, on the basis of 10⁴ Monte Carlo simulations of the best-fit power-law model, this ${\rm{\Delta }}{\chi }^{2}$ improvement occurs by chance in about 12% of simulations, so the evidence for the blackbody component is not significant. Furthermore, this component is not straightforward to interpret; it is much harder, for example, than would be expected for thermal emission from the neutron-star surface (e.g., Rutledge et al. 1999).

The X-ray spectrum also shows marginal evidence for an absorption feature at about 40 keV (Figure 6). An additional Gaussian absorption line (represented by gabs in XSPEC) centered at ${E}_{\mathrm{abs}}={40}_{-3}^{+4}$ keV improves the fit by ${\rm{\Delta }}{\chi }^{2}=-11.3$ (with three additional degrees of freedom). The line width and optical depth are poorly constrained, with 90% upper limits of ${\sigma }_{\mathrm{abs}}\lt 8\,\mathrm{keV}$ and ${\tau }_{\mathrm{abs}}\gt 8$, respectively. Using a similar set of spectral simulations, the significance of this improvement is just under 2%, equivalent to a significance of $\sim 2\sigma$. This absorption line does not have a straightforward interpretation, and at its low significance, we believe it is unlikely to be real.

Zoom In Zoom Out Reset image size

Given the substantial short-term variability (Section 3.3.2), we performed an intensity-resolved spectral analysis to determine whether the photon index varies with X-ray flux. XSELECT was used to resolve the photon events into two roughly equal populations based on a 100 s bin source light curve: a "high-flux" group (total FPMA/FPMB source count $\gt 0.13$ cts s⁻¹) and a "low-flux" group below this limit. The spectra of both groups can still be well-described by an absorbed power law. With N_H free, the high-flux spectrum has a lower N_H and a softer photon index than the low-flux spectrum. We also fit fixed N_H models using the best-fit value of the average spectrum (N_H = 5.8 × ${10}^{22}$ cm⁻²). We found a photon index of ${{\rm{\Gamma }}}_{{\rm{H}}}=1.76\pm 0.06$ in the high state and ${{\rm{\Gamma }}}_{{\rm{L}}}=1.56\pm 0.08$ in the low state, giving a formal difference of $2\sigma$ between the photon indices. Thus these data are consistent with either a variation in N_H over the orbit or with a changing photon index.

All the spectral fitting results are summarized in Table 3.

Table 3.  X-Ray Spectral Fitting Results

Instr.a and Model	CFPMB	CXRT	NH	Γ	${F}_{X,\mathrm{pow}}$ b	Tbb	${F}_{X,\mathrm{bb}}$ b	${\chi }^{2}/\mathrm{dof}$
			(1022 cm−2)		(10−12 erg s−1 cm−2)	(keV)	(10−12 erg s−1 cm−2)
SW (PL)	⋯	⋯	${2}_{-2}^{+4}$	${1.0}_{-1.6}^{+2.1}$	${2.0}_{-0.5}^{+15.9}$	⋯	⋯	2.5/1
NU (PL)	1.05 ± 0.05	⋯	${5.8}_{-1.7}^{+1.8}$	${1.68}_{-0.08}^{+0.09}$	6.7 ± 0.4	⋯	⋯	254.03/280
NU+SW (PL)	1.06 ± 0.05	0.9 ± 0.2	${4.4}_{-1.2}^{+1.6}$	${1.63}_{-0.07}^{+0.08}$	6.8 ± 0.4	⋯	⋯	260.31/283
NU (PL+BB)c	1.06 ± 0.05	⋯	${6.5}_{-2.4}^{+2.8}$	${1.92}_{-0.24}^{+0.37}$	${4.7}_{-1.3}^{+1.6}$	${5.6}_{-1.7}^{+1.8}$	${1.0}_{-0.8}^{+0.9}$	249.77/278
NU (GABS*PL)d	1.06 ± 0.05	⋯	${4.9}_{-1.8}^{+1.9}$	1.61 ± 0.09	7.0 ± 0.5	⋯	⋯	242.70/277
NU high-flux (PL)	${1.05}_{-0.06}^{+0.07}$	⋯	${4.8}_{-2.1}^{+2.3}$	1.72 ± 0.11	11.8 ± 0.09	⋯	⋯	139.84/158
NU low-flux (PL)	1.06 ± 0.08	⋯	${7.7}_{-3.0}^{+3.3}$	${1.63}_{-0.14}^{+0.13}$	${4.5}_{-0.4}^{+0.5}$	⋯	⋯	113.86/142
NU high-flux (PL)	${1.05}_{-0.06}^{+0.07}$	⋯	5.8 (fixed)	1.76 ± 0.06	${11.7}_{-0.08}^{+0.09}$	⋯	⋯	140.33/159
NU low-flux (PL)	1.06 ± 0.08	⋯	5.8 (fixed)	1.56 ± 0.08	${4.6}_{-0.4}^{+0.5}$	⋯	⋯	114.88/143

Notes.

^aSW = Swift; NU = NuSTAR. ^bThe energy range is 0.3–10 keV for Swift/XRT and 3–79 keV for NuSTAR. All fluxes are as measured averaged over the orbit, and should be increased by 6% for the mean out-of-eclipse value. ^cThe significance of the improvement over a simple power-law model is about 1σ. ^dThis model also includes an absorption line with ${E}_{\mathrm{abs}}={40}_{-3}^{+4}$ keV, ${\sigma }_{\mathrm{abs}}\lt 8\,\mathrm{keV}$, and ${\tau }_{\mathrm{abs}}\gt 8$. The significance of the improvement over a simple power-law model is about 2σ.

Download table as:  ASCIITypeset image

The NuSTAR light curve, showing an effective count rate in 500 s bins, is shown in Figure 7. There is strong factor of a few variability on timescales of hundreds of seconds, as well as intervals where the count rate is consistent with zero. The NuSTAR data span about 2.7 cycles of the known 8.8 hr orbital period, though only about 71% of this full time span is on-source. We folded the light curve on the optical ephemeris, using ${P}_{\mathrm{orb}}=31684.61\,{\rm{s}}$ and setting $\phi =0$ at a BJD of 2457527.59971. A 30-bin phased light curve is shown in Figure 8. At $\phi =0.75$, we detect an X-ray dip with a mean count rate of $(3.4\pm 2.1)\times {10}^{-3}$ cts s⁻¹, compared to a mean count rate of $5.4\times {10}^{-2}$ cts s⁻¹ ($\sigma =1.2\times {10}^{-2}$ cts s⁻¹) for the off-dip phases. Given that the X-ray flux formally drops by a factor of ∼16 and is consistent with zero, and that three separate dips are observed at the same phase (Figure 7), we identify these events as X-ray eclipses.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To better constrain the length of the eclipse, we used a finer binning, finding that 500 bins (each of about 63.4 s and, on average, about 10 counts per bin) provided a good balance between time resolution and per bin statistics. We found that the X-ray eclipse has a duration of about 2200 ± 60 s. The depth and duration of each individual eclipse are consistent with the value determined from the phased light curve. Because we detected only about 20 counts during eclipse total, the spectral information is not well-constrained. For the same reason, the non-eclipse X-ray spectrum is essentially identical to the average spectrum presented earlier, with the flux scaled up by 6.0% to correct for the source "off" time.

In addition to the eclipse, there is some evidence for low-frequency modulation in the phased light curve, with a maximum around $\phi \sim 0.1$ and a minimum somewhere in the range of $\phi \sim 0.5$–0.7. To estimate the significance of this variation, we compared the average count rates between $\phi \,=$ 0–0.5 and ϕ = 0.5–1.0 (excluding the eclipse). The difference in count rates is $(1.02\pm 0.17)\times {10}^{-2}$ cts s⁻¹, suggesting evidence for orbital variability (in addition to the eclipse) at over $5\sigma$. Consistent with the spectral analysis presented above, these variations could be due either to changes in N_H over the orbit or with a real change in the X-ray spectrum.

Finally, as a first test for the presence of X-ray pulsations, we divided the NuSTAR data into 10 segments and searched for periodic signals using efsearch, with time steps of 10 ms for periods $\gt 100$ ms and time steps of 0.1 ms for shorter periods. No significant signals were found. However, we emphasize that such searches are extremely challenging and rarely successful unless the radio ephemerides are known. As an example, the X-ray pulsations in the XMM observations of XSS J12270–4859 were not found in an initial search by de Martino et al. (2013) but were discovered once the radio ephemeris were known (Papitto et al. 2015).

We analyzed the PASS 8 Fermi/LAT data of 3FGL J0427.9–6704 during the first 7.5 years since the mission (from 2008 August 04 to 2016 February 20; $\mathrm{ROI}=20^\circ$) to try to improve the source localization and to better constrain the light curve. Using the Fermi Science Tools (v10r0p5), we performed a binned analysis with a source model based on the 3FGL catalog (Acero et al. 2015). All cataloged sources within ${30}^{\circ }$ of our target were included. Spectral parameters of the field sources within 7° from the target and a few variable sources outside 7^◦ were allowed to vary during fitting (for sources at distances of $6^\circ \mbox{--}7^\circ$ and the distance variable sources, only the normalizations were freed). In the course of this analysis, we detected an additional "source" with strong γ-ray emission (Test Statistic (TS) = 188, equivalent to a detection significance of $13.7\sigma$) $1\buildrel{\circ}\over{.} 6$ away from the target, far in the outskirts of the LMC. This could be a true uncataloged source or residual LMC emission. In any case, an extra point source with a power-law spectrum was added to minimize this source of unknown emission.

Using a binned likelihood analysis, we fit the Fermi/LAT spectrum of the 3FGL J0427.9–6704 spectrum with a power law. The best-fit spectral parameters are ${{\rm{\Gamma }}}_{\gamma }=2.48\pm 0.06$ and ${F}_{0.1\mbox{--}300\mathrm{GeV}}=(9.4\pm 0.6)\times {10}^{-12}$ erg s⁻¹ cm⁻² with $\mathrm{TS}\,=441$. These parameters are entirely consistent with that in the 3FGL, which are based on the Pass 7 reprocessed data of the first four years (${{\rm{\Gamma }}}_{\gamma }=2.46\pm 0.07$ and ${F}_{0.1\mbox{--}100\mathrm{GeV}}\,=(9.4\pm 0.8)\times {10}^{-12}$ erg s⁻¹ cm⁻²). This is good evidence that the source did not vary in flux or spectrum over the last approximately four years. The spectral fit is plotted in Figure 9.

Zoom In Zoom Out Reset image size

There is no evidence for curvature in the spectrum, as would be expected for an exponential cutoff: if this is left as a free parameter in the fit, it defaults to the high limit of the range allowed, and the quality of the fit decreases. The spectrum of 3FGL J0427.9–6704 appears to be a pure power law.

We also ran gtfindsrc to optimize the source location based on the best-fit source model. The optimized coordinates are J2000 R.A. and decl. of (04:27:46.23, −67:03:29.6; error radius at 95% confidence of 2'). This positional uncertainty is much improved from 3FGL, and the X-ray and optical sources still sit well within the error circle, providing yet stronger evidence that the γ-ray source is associated with the emission at other wavelengths (Figure 1).

Using the 0.1–300 GeV events extracted with a 1° radius aperture and TEMPO2 (Hobbs et al. 2006) with the Fermi plugin, we made a 10-bin γ-ray phase-folded light curve with phases identical to the X-ray one. This light curve has 4800 counts. The light curve is generally flat with orbital phase with a mean count of 480 ± 22 cts bin⁻¹. However, we also observe a dip to 429 ± 21 cts bin⁻¹ at $\phi =0.7-0.8$, matching the phase of the X-ray eclipse (Figure 8). To check this finding, we performed likelihood analyses on the LAT data in the eclipse ($\phi =0.713-0.783$) and off-eclipse phases. The best-fit γ-ray fluxes are ${F}_{0.1\mbox{--}300\mathrm{GeV}}=(4\pm 2)\times {10}^{-12}$ erg s⁻¹ cm⁻² and ${F}_{0.1\mbox{--}300\mathrm{GeV}}=(9.8\pm 0.6)\times {10}^{-12}$ erg s⁻¹ cm⁻² in and out of eclipse, which formally differ at the $2.8\sigma$ level. Thus we view the evidence for a γ-ray eclipse as suggestive but not yet definitive.

The system distance implied by the best-fit light curve is 2.4 kpc. This value would change by about 10% for the modest variations in secondary mass discussed above, but the systemic uncertainties, such as the metallicity of the system, are much larger. At this distance, the binary would be located about 1.5 kpc below the Galactic plane. This could be consistent with either thick disk membership or with a system that was kicked out of the thin disk.

A separate constraint on the system distance comes from the well-known correlation between a combination of the X-ray luminosity and orbital separation and the optical spectral luminosity density (van Paradijs & McClintock 1994), which is generally interpreted as evidence that the optical luminosity is due to reprocessed X-ray emission. Using the more complete compilation of neutron-star low-mass X-ray binaries in Russell et al. (2007), we find a relatively weak constraint: any distance in the range of ∼0.4–6 kpc would be consistent with the fairly large scatter in the correlation. Nonetheless, this constraint is consistent with the more precise values obtained from the light-curve fitting.

The signpost of transitional millisecond pulsars in the disk state is X-ray "mode switching": rapid, factor of aproximately five variations in the X-ray luminosity sometimes associated with a slightly softer spectrum in the high state (Linares 2014). This behavior has been observed in all three of the confirmed transitional systems (de Martino et al. 2013; Linares et al. 2014; Patruno et al. 2014; Bogdanov et al. 2015) as well as in 1RXS J154439.4–112820, which shows comparable X-ray variability (Bogdanov & Halpern 2015; Bogdanov 2015). In 3FGL J0427.9–6704, we observe rapid X-ray variability, including bins with a count rate consistent with zero outside of eclipse, as well as evidence for a softer spectrum at higher count rates. The clinching evidence for mode switching would be the clear identification of a bimodal count rate distribution, but this is harder to verify for this system than for the others due to our lower count rate and thus requires coarser binning.

The X-ray spectrum and inferred luminosity provide additional evidence that 3FGL J0427.9–6704 is a transitional system. The NuSTAR 3–79 keV eclipse-corrected flux is (7.1 ± 0.4) $\times {10}^{-12}$ erg s⁻¹ cm⁻², giving a luminosity of ($4.9\pm 0.3)\times {10}^{33}{(d/2.4\mathrm{kpc})}^{2}$ erg s⁻¹. To compare to the X-ray properties of the known transitional millisecond pulsars, this luminosity implies 0.5–10 keV luminosity of ($2.4\pm 0.2)\times {10}^{33}{(d/2.4\mathrm{kpc})}^{2}$ erg s⁻¹. In the context of interpreting 3FGL J0427.9–6704 as a candidate transitional millisecond pulsar, we take this luminosity as an average between the high and low modes in the disk state. Given that the known transitional systems have high state disk luminosities of 3–$5\times {10}^{33}$ erg s⁻¹ (0.5–10 keV) and spend ∼60%–80% of the time in the high state (Linares 2014), our observed luminosity of $2.4\times {10}^{33}$ erg s⁻¹ is precisely as expected for a transitional millisecond pulsar in the disk state. The photon index observed (${\rm{\Gamma }}\sim 1.7\pm 0.1$) is consistent with that of M28I, PSR J1023+0038, and 1RXS J154439.4–112820 in their disk states (Linares 2014; Tendulkar et al. 2014; Bogdanov 2015), and the evidence for a softer spectrum at higher count rates is as observed for XSS J12270–4859 (de Martino et al. 2013; Linares 2014). The 8.8 hr orbital period and low-mass main-sequence companion are consistent with the binary properties of the known transitional millisecond pulsars (with orbital periods in the range of 4.8–11.0 hr; Archibald et al. 2009; Papitto et al. 2013; Bassa et al. 2014; de Martino et al. 2014; McConnell et al. 2015).

One possible difference is that 3FGL J0427.9–6704 shows evidence for low-frequency orbital variability in X-rays (Section 3.3.2), while for the known transitional systems, it is only observed in the pulsar state and not for the disk state (Li et al. 2014; Tendulkar et al. 2014; Bogdanov et al. 2015; Bogdanov 2015). This tension can be alleviated if we attribute the orbital variations in 3FGL J0427.9–6704 to variations in N_H due to the high inclination. This is consistent with our observations.

γ-ray emission is emerging as an important aspect to the phenomenology of transitional millisecond pulsars. In the 2013 transition of PSR J1023+0038 from the pulsar to disk state, the γ-ray emission increased by a factor of six (Stappers et al. 2014; Takata et al. 2014; Deller et al. 2015), and in XSS J12270–4859 the γ-ray emission decreased by a factor of two as it transitioned out of the disk state (Johnson et al. 2015). No variations have been detected for M28I, but this is non-informative due to its location among other γ-ray sources in a globular cluster and larger distance.

Excluding M28I, the ratio of 0.1–300 GeV γ-ray luminosity to 0.5–10 keV X-ray luminosity for the three other systems lie in the range of two to four (Linares 2014; Bogdanov & Halpern 2015; Deller et al. 2015; Johnson et al. 2015). From our Fermi analysis of 3FGL J0427.9–6704, the 0.1–300 GeV flux of the source is (9.4 ± 0.8) × ${10}^{-12}$ erg s⁻¹ cm⁻², equivalent to a luminosity of ($6.5\pm 0.6)\times {10}^{33}{(d/2.4\mathrm{kpc})}^{2}$ erg s⁻¹. This gives a γ-ray to X-ray ratio of ∼2.7, right in line with the other transitional millisecond pulsars.

There are possibly interesting differences among the γ-ray spectra of these systems. For both XSS J12270–4859 and PSR J1023+0038, the γ-ray spectrum appears best fit by a power law with an exponential cutoff in the disk state (Takata et al. 2014; Johnson et al. 2015); for 3FGL J1544.6–1125 the 3FGL catalog reports a power-law fit with ${\rm{\Gamma }}=2.36\pm 0.08$ (Acero et al. 2015). We find that the Fermi spectrum of 3FGL J0427.9–6704 is well-fit by a power law with ${\rm{\Gamma }}=2.48\pm 0.06$ and that a single power law is strongly preferred over a model with an exponential cutoff.

We note that the γ-ray spectrum of 3FGL J0427.9–6704 is dissimilar to that of pulsars detected with Fermi/LAT, which tend to have a shallower (${\rm{\Gamma }}\lt 2$) power-law-like spectrum at lower energies and a sub-exponential cutoff at a few GeV. Hard power-law spectra are observed for pulsar wind nebulae, in which the relativistic wind of a young pulsar interacts with surrounding material, producing high-energy photons through synchrotron and inverse Compton emission (Abdo et al. 2013).

This simple power-law spectrum of 3FGL J0427.9–6704 is reminiscent of the "huntsman"⁴ system 1FGL J1417.7–4407, which has a comparable γ-ray photon index with no evidence for curvature (Strader et al. 2015). This system shows both similarities and differences to the known transitional systems in the disk states: it has a hard power-law X-ray spectrum and double-peaked Hα that suggests an accretion disk, but also has a giant companion in a longer period rather than a main-sequence companion, a higher ratio of γ-ray to X-ray luminosity (∼20), and was visible as a radio pulsar at similar times to the optical observations (Camilo et al. 2016). One interpretation of 1FGL J1417.7–4407 is that despite the presence of a disk, all of the accretion flow is being ejected in a propeller and thus none reaches the neutron-star surface to halt the pulsar mechanism.

The other notable feature of the γ-ray emission from 3FGL J0427.9–6704 is the tentative evidence for a γ-ray eclipse. Taken at face value, our measurements suggest a partial eclipse, indicating the γ-ray emission comes from a large region that is not entirely blocked by the secondary during the X-ray eclipse. At present, the evidence for an eclipse is just under $3\sigma$. The measured flux during the eclipse is only $2\sigma$ above zero, and so the possibility of a total eclipse, which would imply a smaller γ-ray production region, cannot be excluded. We note that Romani et al. (2015) have found tentative evidence for the eclipse of the magnetospheric γ-ray emission in the non-accreting millisecond pulsar binary PSR J2215+5135.

As many of recent papers cited above have extensive discussions of the physical models for transitional millisecond pulsars, a detailed recapitulation here is superfluous. Considering the main scenarios for γ-ray production in the disk state: one possibility is that the γ-rays originate either in a strong shock between the pulsar wind and surrounding material associated with the accretion flow (e.g., Stappers et al. 2014) or due to inverse Compton scattering of disk photons off the pulsar wind (e.g., Takata et al. 2014). These scenarios are disfavored by the recent observations of coherent X-ray pulsations in the disk states of PSR J1023+0038 and XSS J12270–4859 (Archibald et al. 2015; Papitto et al. 2015). The material reaching the surface would be expected to quench the pulsar wind. An alternative scenario is synchrotron self-Compton at the interface between the pulsar magnetosphere and the accretion disk (e.g., Papitto et al. 2014; Deller et al. 2015). For this latter model, the γ-rays would originate in a small region close to the neutron star, so a total (rather than a partial) eclipse would be expected.

It is clear that the details of the γ-ray eclipse in 3FGL J0427.9–6704 represent important constraints on the γ-ray production mechanism that can be improved with future Fermi observations.

By definition, confirmation of a system as a transitional millisecond pulsar can only come via an observed transition between the disk and pulsar states. For 3FGL J0427.9–6704, the OGLE photometry implies the existence of a similar accretion disk since 2010, so the system has been visible as a low-mass X-ray binary for at least six years. The constancy of the Fermi flux over a similar timescale (Section 3.3.3) is additional evidence for no transition during this time. Thus, as for the case with 1RXS J154439.4–112820 (which has been in the disk state for at least 10 years; Bogdanov 2015) we are in somewhat of a waiting game to see whether a transition will occur. In any case, the discovery of these systems shows that we are far from complete in our census of γ-ray bright compact binaries associated with Fermi sources.