DISCOVERY OF THE NEAR-INFRARED COUNTERPART TO THE LUMINOUS NEUTRON-STAR LOW-MASS X-RAY BINARY GX 3+1^*

Low-mass X-ray binaries (LMXBs) are systems in which a neutron-star or black hole accretes matter from a low-mass secondary (M ≲ 1 M_☉). Most are only occasionally bright in X-rays, but some have been very X-ray luminous since their discovery. Neutron-star low-mass X-ray binaries (NS-LMXBs) with relatively weak magnetic fields show a wide variety of correlated spectral and variability behavior in X-rays. Based on this, two main subclasses are recognized (Hasinger & van der Klis 1989): the so-called Z sources, with luminosities close to or above the Eddington luminosity (L_Edd), and the atoll sources, with luminosities up to ∼0.5 L_Edd; see Homan et al. (2010) for an overview.

To explain the high mass-accretion rates (∼10⁻⁹–10⁻⁸ M_☉ yr⁻¹) that are implied by their X-ray luminosities, it has been suggested that the mass donors in the most luminous (L_X ≳ 1 × 10³⁷ erg s⁻¹) NS-LMXBs are evolved stars (Webbink et al. 1983; Taam 1983). The evolutionary expansion of (sub-)giant companions can drive mass transfer rates in excess of those that occur in systems with main-sequence donors, where mass-transfer is driven by the loss of angular momentum. The evolutionary stage of the donor also affects the duration of the active X-ray–binary phase in the lifetime of an LMXB and might shape the LMXB X-ray luminosity function (XLF). Revnivtsev et al. (2011) suggested that the break around log L_X(erg s⁻¹) ≈ 10^37.3 in the LMXB XLF of our own Galaxy and nearby galaxies (Gilfanov 2004) separates systems with main-sequence companions from those with giant donors whose shorter active lifetimes would explain the steepening of the XLF. Their argument uses orbital period as a proxy of donor size, and observations of a handful of systems confirm that this is a valid assumption in those cases. However, for about half of the sources with L_X > 10^37.3 erg s⁻¹ an orbital period or donor classification is not available.

Optical and near-infrared (OIR) studies can potentially clarify some of these issues. They shed light on the structure of the various accretion flow components, such as disks and jets (Russell et al. 2007) and can provide information on the binary parameters, such as orbital period and properties of the mass donor (e.g., Bandyopadhyay et al. 1999). Given that the population of bright NS-LMXBs is concentrated toward the heavily obscured Galactic bulge, the low extinction in the near-infrared (NIR)—A_K ≈ 0.1 A_V—makes it the preferred band to carry out such studies in that region.

With a luminosity of (2–4) × 10³⁷ erg s⁻¹ (2–10 keV; den Hartog et al. 2003) GX 3+1 is one of the eleven most luminous and persistently bright Galactic NS-LMXBs. Ever since its discovery in 1964 (Bowyer et al. 1965) observations of GX 3+1 have found it to be a bright X-ray source. The detection of X-ray bursts (Makishima et al. 1983) indicates that it is a neutron star system. The best distance estimate of ∼6.1 kpc is derived from the properties of a radius-expansion burst, assuming an Eddington limit that is appropriate for a hydrogen-poor atmosphere (Kuulkers & van der Klis 2000; den Hartog et al. 2003). The resulting maximum persistent bolometric luminosity is ∼6 × 10³⁷ erg s⁻¹ or ∼0.3 L_Edd. Based on its spectral and variability properties, GX 3+1 is classified as an atoll source (Hasinger & van der Klis 1989). Similar to the bright atolls GX 9+1 and GX 9+9, GX 3+1 shows strong long-term X-ray flux modulations, which have a timescale of ∼6 yr (Kotze & Charles 2010; Durant et al. 2010). Kotze & Charles (2010) suggested these modulations are the result of variations in the mass-transfer rate due to a solar-type magnetic-activity cycle in the donor star. However, since the long-term X-ray light curve of GX 3+1 covers only ∼2.5 cycles of the brightness modulations, it is unclear if they are quasi-periodic or have a more random nature and are simply a manifestation of very-low-frequency noise. The lack of an accurate position for GX 3+1 has hampered the search for a counterpart, and thus far the orbital period of GX 3+1 remains unknown.

Naylor et al. (1991) were the first to look for an NIR counterpart to GX 3+1. They identified one candidate inside the region defined by the Einstein error circle and the lunar-occultation error box. More recently, Zolotukhin & Revnivtsev (2011) repeated the search using an additional position constraint imposed by a Chandra observation performed in continuous-clocking mode. In the highly elongated region of about 05 × 2'', they identified two possible counterparts, viz. the one originally found by Naylor et al. (1991) and a much fainter source.

Here, we present the results of our own search for GX 3+1 in the NIR. Combining a newly determined accurate X-ray position with NIR imaging and spectroscopy, we have uncovered the true counterpart of GX 3+1 five decades after its discovery in X-rays. We describe the X-ray and NIR observations in Section 2, and the identification of the counterpart in Section 3. In Section 4, we discuss the origin of the NIR emission in GX 3+1, and make a comparison with other NS-LMXBs.

2.1. X-Rays

2.1.1. Chandra HRC Observation

We observed GX 3+1 with the Chandra High Resolution Camera imaging detector (HRC-I; Zombeck et al. 1995) for 1.2 ks, starting on 2012 February 12 at 19:22:43 UTC. The High Energy Transmission Grating was inserted to lower the count rate and to limit distortions in the image that could adversely affect the image reconstruction. The data were analyzed using CIAO 4.5 (Fruscione et al. 2006). We ran the wavdetect tool on a 200 × 200 pixel image, centered on GX 3+1. This yielded the following source position (R.A., decl.): 17^h47^m56077, −26°33'4948 (J2000); the formal wavdetect errors are negligible (<001 in each coordinate). About 2.1 × 10⁴ photons were detected in a 2'' radius circle centered around this position. No other X-ray sources were detected in the field, which made it impossible to improve the absolute astrometry of the image using X-ray sources with known accurate (radio or optical/NIR) counterpart positions. Therefore, the positional accuracy is limited by Chandra's absolute astrometric calibration, which for the HRC-I is ∼054 (90% confidence).⁶

2.1.2. RXTE/ASM and MAXI Light Curves

We used public data from the All Sky Monitor (ASM) on board the Rossi X-ray Timing Explorer (RXTE; Levine et al. 1996) and from the Monitor of All-sky X-ray Image (MAXI) camera (Matsuoka et al. 2009) to construct a long-term 2–10 keV light curve of GX 3+1 from one-day average count rates (ctr). Data points with large uncertainties (ctr/σ_ctr < 20, with σ_ctr the standard deviation in ctr) were removed; for the MAXI data, we also removed a few-day-long flare around MJD 55916, whose origin and relation to GX 3+1 are unclear, and data between MJD 56187 and MJD 56351, which were affected by an outburst from a transient in the globular cluster Terzan 5. The count rates from both instruments were normalized to those of the Crab. The resulting light curve is shown in Figure 1. Strong long-term modulations with time-scales of several years are clearly visible. The vertical lines in Figure 1 indicate the times of our NIR imaging observations (see next section); the first two were only separated by about one day and appear as a single line.

Zoom In Zoom Out Reset image size

2.2. Near-infrared

Figure 2 shows the region around the new Chandra position of GX 3+1 in the K_s images from all four imaging epochs. The red circles represent the 90% confidence radius on the source position. These were computed by adding in quadrature the Chandra absolute pointing error (054) and the errors in the astrometric solutions of the PANIC and FourStar images (see Sections 2.2.1 and 2.2.2) scaled to 90% errors assuming a two-dimensional Gaussian distribution. For illustrative purposes, we also show the combined 95% error radius in Figure 2(c).⁷ From this figure, which shows the image taken under the best seeing conditions, it is clear that multiple K_s-band sources are viable counterparts to GX 3+1 based on their locations inside or very close to the 90% (058 for this epoch) or 95% (083) confidence radii; in principle, any of the sources marked with a small yellow circle could be the counterpart. Comparison of this image with the lower resolution UKIRT K image in Figure 5 of Zolotukhin & Revnivtsev (2011) shows that their candidate counterpart A is now resolved into four sources, which we label A 1 to A 4 in order of decreasing K_s brightness. Source B, the other counterpart proposed by Zolotukhin & Revnivtsev (2011), lies too far from the new Chandra position to still be considered a possible counterpart. Table 2 lists the coordinates of these stars together with their K_s magnitudes from epochs 2 and 3, during which they are detected without significantly suffering from blending.

Zoom In Zoom Out Reset image size

Table 2. Positions and Magnitudes of NIR Sources Near the New Position of GX 3+1

ID	αJ2000	δJ2000	Δ	Ks, epoch2	Ks, epoch3
			('')
A 1	17h47m56063	−26°33'4874	0.76	15.62 ± 0.01	15.62 ± 0.01
A 2	17h47m56097	−26°33'4935	0.31	15.82 ± 0.01	15.84 ± 0.01
A 3	17h47m56015	−26°33'4906	0.87	16.78 ± 0.02	16.84 ± 0.02
A 4	17h47m56056	−26°33'4971	0.37	18.12 ± 0.05	18.11 ± 0.04
B	17h47m56214	−26°33'4909	1.94	17.26 ± 0.02	17.13 ± 0.02

Notes. Δ is the offset between the positions of the NIR sources and the new Chandra position of GX 3+1. We consider A 2 to be the likely NIR counterpart to GX 3+1 based on the detection of a Br γ emission line in its K_s-band spectrum. The uncertainties in the magnitudes are the DAOPHOT errors on the PSF photometry. Additional errors in the photometric calibration with respect to 2MASS are <0.1 mag and are given in Sections 2.2.1 and 2.2.2.

Download table as:  ASCIITypeset image

The two candidate counterparts for which we obtained FIRE spectra are A 1 and A 2, which lie just outside and inside the 90% error circle, respectively. Whereas the spectrum of A 1 is featureless, the spectrum of A 2 clearly shows Br γ in emission (Figure 3). This H i emission feature is commonly associated with accreting sources (Bandyopadhyay et al. 1999, 2003), and its presence shows that A 2 is the true NIR counterpart of GX 3+1. Other H i, and possibly He i and He ii, lines may be identified as well, but are much weaker; the 2.192 μm feature especially is very weak and may turn out not to be a real emission line in a higher signal-to-noise spectrum. Nevertheless, the spectrum of A 1 does not show similar features at the corresponding wavelengths. In Table 3, we list the measured wavelengths of the (tentative) emission lines. The equivalent width of the Br γ line is about −45 ± 5 Å; given that the spectrum of A 2 still contains a small contribution from the light of A 1, the true value must be lower (i.e., more negative). No absorption features from a companion star can be seen in the spectrum of A 2 but we point out the poor signal-to-noise ratio especially in the H band (due to poor sky subtraction) and beyond ∼2.3 μm, where the CO absorption bands appear as prominent features in the K-band spectra of stars of spectral type K and M.

Zoom In Zoom Out Reset image size

We compare the magnitudes for star A 2 from epochs 2 and 3 to check for variability. Taking into account both the formal photometry errors from DAOPHOT and the photometric calibration errors, we do not see any sign of variability in K_s in excess of ∼0.1 mag in our own observations (Table 2). The combined magnitudes of A 1 to A 4 are consistent with K = 14.87  ±  0.13 mag for source A on 2007 May 3 as reported by Zolotukhin & Revnivtsev (2011), and with K = 15.13  ±  0.16 mag for that same source (i.e., star 311) on 1988 May 11–15 as reported by Naylor et al. (1991). This points to a lack of large-amplitude NIR variability on a timescale of years.

We also searched the Vista Variables in the Via Lactea (VVV) catalogs (Minniti et al. 2010) for detections of A 2. The most recent data release (DR), i.e., DR 3, has 18 images that cover the position of GX 3+1, including 10 epochs in the K_s band. However, in these images A 2 is not detected as a separate object but as a blend with, at least, A 1, or with A 1 and the bright object to the southwest of A 4 just outside of the 95% error circle (see Figure 2(c)). The poorer image quality of the VVV images compared to that of our images from epochs 2 and 3 is due partly to the coarser VVV pixel scale (034 pixel⁻¹) and partly to the seeing conditions under which these VVV images were taken (≳08). Reprocessing of the VVV images to attempt to extract deblended magnitudes for A 2 and A 1 is beyond the scope of this paper. A visual examination of the DR 3 images does not reveal any obvious brightness variations of A 2. The recently released multi-band master catalog extracted from the DR 1 images⁸ did not include a detection of A 2, either.

The NIR emission from bright NS-LMXBs can originate in several parts of the system: the accretion disk may contribute via thermal emission that results from X-ray or viscous heating, a late-type secondary can produce thermal emission in the NIR, which may be enhanced by X-ray heating of the hemisphere facing the accretion region, and finally the inner region of a jet—if present—may contribute through optically thin synchrotron emission. Spitzer data of the NS-LMXB 4U 0614+091 suggest that one would have to observe at wavelengths >8 μm to detect emission from the optically thick part of the jet (Migliari et al. 2010). For jets that are significantly stronger than the one in 4U 0614+091 the break between the optically thin and thick part of the jet may shift to shorter wavelengths (Falcke et al. 2004).

The FIRE spectrum of GX 3+1 indicates that at least a significant portion of the NIR light comes from a heated accretion disk or heated secondary. Both the emission lines and the blue (i.e., F_ν∝ν^α with α > 0) continuum of the unreddened spectrum (Figure 4) are as expected for a thermal component (Russell et al. 2007). Based on our NIR data alone, we are not able to distinguish between thermal disk emission resulting from either X-ray or viscous heating. This would require simultaneous NIR and X-ray observations to look for possible correlated behavior on timescales of seconds, as would be predicted by X-ray heating.

Zoom In Zoom Out Reset image size

GX 3+1 has not been detected at radio wavelengths (Berendsen et al. 2000), suggesting there is no powerful jet in this system. Furthermore, optically thin synchrotron emission produces a red (α < 0) NIR spectrum, which is clearly inconsistent with the spectrum in Figure 4. We note that the exact slope of the continuum is uncertain as contamination by the light from the close neighbor A 1 is not completely accounted for (see Section 2.2.3). However, in Figure 3, one can see that the observed spectrum of A 2 is bluer than that of A 1, thus, if anything, the effect of A 1 is to make the spectrum of A 2 seem redder. We conclude that most likely a jet does not significantly contribute to the NIR emission of GX 3+1.

To estimate the possible contribution of an unheated secondary, we first use the estimate of the column density toward GX 3+1 from Oosterbroek et al. (2001), viz. $N_{\rm H}=1.59^{+0.07}_{-0.12}\times 10^{22}$ cm⁻², and the distance of 6.1 kpc (with an estimated uncertainty of ∼15%; Kuulkers et al. 2003) to compute the absolute K_s magnitude of A 2. Adopting the relation N_H = 1.79 × 10²¹ A_V cm⁻² from Predehl & Schmitt (1995), A_K = 0.114 × A_V from Cardelli et al. (1989), and $A_K=0.95 \times A_{K_s}$ from Dutra et al. (2002), we find $M_{K_s}=0.84 \,{\pm}\, 0.35$, where the error is dominated by the uncertainty in the distance. However, since GX 3+1 lies close to the direction of the Galactic center, it may be more appropriate to adopt the conversion $A_{K_s}=0.062 \times A_{V}$ from Nishiyama et al. (2008), which yields $M_{K_s}=1.35\pm 0.34$. We compare this value to the absolute K_s magnitudes of late-type dwarfs and giants, which typically are the donors in LMXBs. Main-sequence stars of spectral type G0 to M5 have $M_{K_s} = 3.0$ to 8.4 (Mamajek 2014.⁹) Therefore, if the secondary is indeed a late-type dwarf, it does not provide a dominant (>50%) contribution to the K_s flux. Giants of spectral type G0 and later have $M_{K_s}\lesssim -0.45$ (Ostlie & Carroll 2007; Bessell & Brett 1988) which is brighter than the bright limit to the estimated $M_{K_s}$ for GX 3+1; therefore, we can exclude that the secondary is a late-type giant.

Given that the K_s flux from GX 3+1 is most likely thermal, we can estimate the orbital period, P_b, using $M_{K_s}$ and the X-ray luminosity, L_X. A relation between these properties is expected if the NIR emission originates from thermal emission in the outer parts of the accretion disk (where X-ray reprocessing dominates over viscous heating), the size of which is set by the orbital separation of the binary stars—and thus the orbital period. van Paradijs & McClintock (1994) demonstrated the existence of such a correlation in the optical, and it has recently been extended to the NIR by Revnivtsev et al. (2012) based on a small sample of NS-LMXBs. For L_X/L_Edd = 0.1–0.2 and $M_{K_s}=1.35\pm 0.34$, we find that P_b ≈ 4.4–10.9 hr using the equation in Section 4.2 in Revnivtsev et al. (2012)¹⁰; for $M_{K_s}=0.84\pm 0.35$, we find P_b ≈ 7.3–18.5 hr. This range of periods also suggests that the secondary is not a giant (see, e.g., Verbunt 1993), though it leaves room for a somewhat evolved star or subgiant. Now that we have identified the NIR counterpart of GX 3+1, a targeted follow-up campaign of time-resolved spectroscopy or photometry may be attempted to directly measure the orbital period.

Russell et al. (2007) investigated the origin of the OIR emission in NS-LMXBs based on a set of near-simultaneous X-ray and OIR data. They found that both at luminosities below L_X ≈ 10³⁶ erg s⁻¹ and at the high X-ray luminosity end (i.e., in the Z sources) most of the NIR emission comes from an X-ray-heated disk; for atolls and milli-second X-ray pulsars of intermediate luminosity, the jet emission dominates. The sample of NS-LMXBs studied by Russell et al. did not include the brightest atolls. In fact, NIR counterparts were discovered only recently for several of these systems, such as GX 9+1 (Curran et al. 2011), 4U 1705−44 (Homan et al. 2009), and now GX 3+1 (this work). The counterpart of GX 9+1 has not been studied in detail, yet. From JHK_s colors and correlations between near-simultaneous X-ray (3–100 keV) and K_s-band data, Homan et al. (2009) found that in 4U 1705−44, a source that can reach similar X-ray luminosities as GX 3+1, the NIR flux is likely the result of X-ray heating, like in the Z sources. They suggested that the high X-ray luminosity could be a sign of a larger disk compared to the disks in fainter atolls, which would explain the larger NIR contribution from X-ray heating. Our findings indicate that GX 3+1 behaves similarly to 4U 1705−44, in the sense that jet NIR emission does not play a (dominant) role; therefore, this scenario could offer a plausible explanation for the origin of the NIR emission in GX 3+1 as well.

Finally, we comment on the lack of variability of the NIR counterpart to GX 3+1. In Figure 1, the quasi-periodic modulations in the RXTE/ASM and MAXI count rate for GX 3+1 are clearly visible. If these brightness variations are truly the result of changes in the mass-accretion rate, it is expected that the NIR emission, which, as we show, contains a dominant contribution from an accretion disk or heated secondary, should also vary in time. For example, in the case of 4U 1705-44, Homan et al. (2009) clearly see correlated X-ray and NIR variations. The difference in X-ray count rate between the epochs of our imaging observations is a factor of ∼2, but surprisingly, we see no change in the K_s magnitude within the photometric accuracy (∼0.1 mag). We note that we only have good measurements of the NIR brightness at two epochs, and need further monitoring to investigate the (lack of) NIR variability of GX 3+1 in more detail.

The authors thank A. Monson for help with the FourStar data reduction and R. Remillard, P. Sullivan, and M. Matejek for obtaining part of the observations. This work is supported by Chandra grant GO2-13048X.

Facilities: CXO - Chandra X-ray Observatory satellite, RXTE - Rossi X-ray Timing Explorer, MAXI - JAXA Monitor of All-sky X-ray Image experiment, Magellan:Baade (PANIC, FourStar, FIRE) - Magellan I Walter Baade Telescope