A Massive AGB Donor in Scutum X-1: Identification of the First Mira Variable in an X-Ray Binary

Scutum X-1 was identified from a sounding rocket experiment nearly 50 yr ago (Hill et al. 1974) as an unusual and bright X-ray source exhibiting the strongest absorption ever detected at the time. The source has been subsequently detected by several X-ray missions (see Kaplan et al. 2007 for a review) and has been noted for exhibiting a conspicuous ≈112 s X-ray pulsation arising from a neutron star (NS) accretor (Koyama et al. 1991). Kaplan et al. (2007) presented the first accurate (to ≈1'') X-ray localization using XMM-Newton observations, which positioned the X-ray source to be within 02 of a bright infrared counterpart (K_s ≈ 6.5 mag) in the 2MASS catalog. The counterpart was reported to be nearly invisible in the optical bands (r > 25.2 mag, i ≈ 23.6 mag; Kaplan et al. 2007). Based on the extremely red colors (J − K ≈ 5.5 mag), HK-band spectroscopy, and observed X-ray absorption, they suggested the counterpart to likely be a late-type supergiant with a two-dimensional classification of ≈M0-1 Ia-Iab. The source was thus classified as a symbiotic X-ray binary (SyXRB; Masetti et al. 2006) at a distance of 4–10 kpc and behind an extinction column of A_V ≳ 25 mag.

The suggested classification makes Sct X-1 the first known candidate for an NS accreting from the wind of a red supergiant (RSG) in a Galactic X-ray binary ¹¹ and an evolved member of the high-mass X-ray binary (HMXB) population. The source still remains one of only three such known systems—the other two sources being the Galactic center X-ray binary CXO 174528.79–290942.8 (Gottlieb et al. 2020) and 4U 1954+31 for which the donor classification was only recently revised from an M4-5 III giant (Masetti et al. 2006) to an M4 I supergiant (Hinkle et al. 2020). Unlike the more common B-supergiant HMXBs (Martínez-Núñez et al. 2017), RSG HMXBs are rare due to their short lifetimes but are of interest as ubiquitous evolutionary phases in the path from massive binaries to double compact objects (Mondal et al.2020).

As noted in Kaplan et al. (2007), the spectrum used to classify the donor in Sct X-1 has uncertain flux calibration as well as limited wavelength coverage. In addition, the high reddening toward the source precludes any attempts at optical spectroscopy or photometry to constrain the donor type via spectral features or variability. Concomitantly, the Gaia catalog does not provide a parallax to constrain its distance and luminosity. The emergence of long-term infrared time-domain surveys combined with moderate-aperture infrared spectroscopy offers new opportunities to revisit the demographics of obscured Galactic X-ray sources such as Sct X-1.

In this paper, we present the first broadband infrared spectrum and near-infrared (NIR) light curve of the infrared counterpart of Sct X-1. Using the observed strong molecular absorption features and large-amplitude photometric variability, we show that Sct X-1 has been long misclassified as an RSG X-ray binary and instead contains a very late-type pulsating Mira donor star. We present the observational data in Section 2. Section 3 presents an analysis of the infrared spectral features and photometric variability to constrain the donor type, distance, and extinction. We discuss the implications of these results and summarize our findings in Section 4.

2.1. Infrared Photometry

Kaplan et al. (2007) showed that the source 2MASS J18352582–0736501 (hereafter 2MASSJ1835) was very likely (chance probability ∼2 × 10⁻⁶) the infrared counterpart of Sct X-1, with a 2MASS magnitude of J ≈ 12 mag. The sky location was monitored as part of regular survey operations of the Palomar Gattini-IR NIR time-domain survey (Moore & Kasliwal 2019; De et al. 2020). Palomar Gattini-IR is a wide-field (≈25 deg²) and shallow J-band survey (reaching J ≈ 13 mag and J ≈ 15 mag ¹² in and outside the Galactic plane, respectively), monitoring the entire visible sky from Palomar Observatory using a 30 cm telescope with a cadence of ≈2 nights. The light curve was obtained by performing forced aperture photometry at the location of the infrared counterpart, with an aperture size of 83. The light curve is shown in Figure 1 and was binned over ≈7 days intervals to improve the signal-to-noise ratio.

Zoom In Zoom Out Reset image size

2.2. Infrared Spectroscopy

We obtained an infrared spectrum of 2MASSJ1835 using SpeX on the NASA Infrared Telescope Facility (Rayner et al. 2003; Program 2021A083, PI: De) on UT 2021 May 16. The data were obtained with a 03 slit, in both the SXD and LXD_long modes over multiple dithered exposures amounting to a total exposure time of ≈720 s and ≈300 s, respectively. The standard stars HIP 90967 and HIP 87108 were used for telluric and flux calibration in the SXD and LXD_long modes, respectively. The combined spectrum covers the full wavelength range of the instrument from ≈0.8 to ≈5.0 μm with a resolution of R ≈ 2000. The data were reduced using the spextool pipeline (Cushing et al. 2004), while telluric and flux calibration was performed using the xtellcor package (Vacca et al. 2003). Due to the high reddening, the source is not detected at <1.0 μm in the final spectrum shown in Figures 2 and 3 (observation epoch is shown with a red arrow in Figure 1). The average signal-to-noise ratio is ≈300 in the K band, ≈200 in the H band, and ≈70 in the J band.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.3. X-Ray Observations

3.1. Improved Localization and Recent X-Ray Flux

3.2. Infrared Spectrum

3.3. Infrared Light Curve

3.3.1. Periodicity Analysis

The well-sampled J-band light curve of 2MASSJ1835 is shown in Figure 1. The light curve shows a clear periodic trend superimposed by smaller-amplitude high-frequency oscillations. Noting the strong evidence for water absorption in the counterpart spectrum (see Section 3.2), we interpret the variability as large-amplitude pulsations of a Mira variable (Kholopov et al. 1985). Mira variables are known to be not strictly periodic but characterized by both low-frequency and high-frequency variations associated with variations in circumstellar dust and supergranular convection (Groenewegen et al. 2007; Iwanek et al. 2021). We use the semiparametric Gaussian process regression model proposed by He et al. (2016) to account for these variations, decomposing the light curve into four parts:

$$\begin{eqnarray}&&g(t)=m+l(t)+q(t)+h(t),\end{eqnarray} \tag{ 1 }$$

where m is the mean magnitude, l(t) is the slowly variable mean magnitude, q(t) is the primary periodic term, and h(t) is the high-frequency stochastic variability. The terms are modeled Gaussian processes with different kernels to account for effects such as changing amplitudes across cycles.

We use the R code distributed by He et al. (2016) to derive a primary pulsation period from the log-likelihood periodogram and a best-fit J-band light-curve model including all the variability terms. We estimate the statistical uncertainty in the period by simulating 100 realizations of the observed light curve by adding Gaussian noise scaled to the photometric uncertainties. In order to estimate the systematic uncertainty given the relatively short baseline (≈1000 days) of our light curve, we selected two Mira variables (OGLE-BLG-LPV-180685 and GLE-LMC-LPV-00055) with similar light curves to Sct X-1 from the OGLE-III database (Udalski et al. 2008), scaled their well-known periods to 621 days, and then examined the scatter of periods found from about fifty 1000 day chunks uniformly distributed along the full light curves. The standard deviation of the periods was 28 days for OGLE-BLG-LPV-180685 and 36 days for OGLE-LMC-LPV-00055, and we thus take the larger as an estimate of the uncertainty. The best-fit light curve and period are shown in Figure 1. The observed peak-to-peak amplitude of ΔJ ≈ 3.5 mag is consistent with the large-amplitude NIR variability seen in long-period symbiotic Miras (Gromadzki et al. 2009). We discuss the implications of the observed pulsation amplitude on the reported archival fluxes of 2MASS J1835 in the Appendix.

3.3.2. Distance and Extinction

Mira variables exhibit a period–luminosity–color relation over several decades in luminosity (Feast et al. 1989; Wood 2000), and that is commonly used to estimate distances to symbiotic Miras (Whitelock 1987; Gromadzki et al. 2009). Using the Whitelock et al. (2008) period–luminosity relation for O-rich Miras suggests an intrinsic absolute magnitude of M_K = −8.6 ± 0.3 mag (M_bol ≈ −5.1). Our inferred NIR luminosity is marginally fainter than the supergiant luminosity (M_K ≈ −9.4) class assumed in Kaplan et al. (2007). Using the period–color relation from Whitelock et al. (2000), we estimate an intrinsic NIR color of (J − K)₀ = 1.59 ± 0.26 mag. The 2MASS J magnitude of the source is consistent with the long-term average observed in the PGIR light curve, and hence we use the 2MASS measurements to infer a E(J − K) = 3.92 ± 0.27 mag.

The estimated color excess corresponds to an optical extinction of A_V = 21.8 ± 1.5 mag (A_K = 2.4 ± 0.2 mag) using the same extinction law as in Kaplan et al. (2007). The estimate corroborates the previous suggestion that Sct X-1 lies behind a large column of dust. The observed K-band magnitude and inferred extinction place the source at a distance of ${3.6}_{-0.7}^{+0.8}$ kpc. The estimated distance is consistent with that estimated by Kaplan et al. (2007), due to the inferred K-band luminosity difference being compensated by the redder intrinsic color of the Mira. Applying the inferred extinction (we assume a Cardelli et al. 1989 extinction law with R_V = 3.1) to the observed 2MASS photometry and fitting a Planck function, we derive an NIR color temperature of ≈2000 ± 190 K, consistent with the effective temperatures of very late-type Mira variables (Haniff et al. 1995; Feast 1996). Overall, the long pulsation period and large amplitude are consistent with a very late-type Mira variable as inferred from the spectral analysis (Feast et al. 1989; Wood & Sebo 1996).

We have presented spectroscopic and photometric evidence to suggest that the donor star in Sct X-1 is an M8-9 III–type pulsating Mira, revising the classification of Kaplan et al. (2007), who favored an RSG donor. Most SyXRB donors have been previously classified with early M-type red giant donors, similar to the first classified source GX 1+4 (Davidsen et al. 1977; Chakrabarty & Roche 1997) that was suggested to be at the tip of the first-ascent red giant branch. Using an optical spectrum, Smith et al. (2012) classified the very red donor of the symbiotic fast X-ray transient XTE J1743–363 as a possible >M7 AGB star; however, the donor photometric variability remains unconstrained. CXOGBS J173620.2–293338 and CGCS 5926 were also proposed as candidate SyXRBs spatially coincident with carbon stars (Masetti et al. 2011; Hynes et al. 2014), but their low X-ray luminosity does not exclude white dwarf (WD) accretors.

Sct X-1 thus represents the first confirmed Mira donor star in an X-ray binary, akin to the relatively rare class of D-type symbiotic stars hosting WDs (Allen & Glass 1974). Our results are consistent with the classification scheme of Akras et al. (2019a, 2019b), who classified Sct X-1 as a likely D-type symbiotic binary based on its IR colors. However, the color-based classification is not completely reliable because they did not account for interstellar extinction, which is a substantial correction in the case of Sct X-1. 4U 1954+31 remains the only confirmed Galactic RSG SyXRB (Hinkle et al. 2020), noting that the donor of CXO 174528.79SyXRB290942.8 remains to be spectroscopically confirmed; Gottlieb et al. (2020) could not exclude the possibility of an AGB donor based on the NIR colors.

Our identification of a ≈621 day pulsation period in 2MASSJ1835 makes the donor one of the longest-period symbiotic Miras known in the Galaxy (Gromadzki et al. 2009). Long-period Miras (>500 days) are known to be relatively rare; their rarity reflects the brevity of the long-period pulsation phase for lower-mass AGB stars or the rarity of higher-mass AGB stars that spend most of their thermally pulsing phase at longer periods (Vassiliadis & Wood 1993; Marigo et al. 2017; Trabucchi et al. 2019). We can thus constrain the evolutionary stage of the system by comparing the observed pulsation period and surface composition (O rich; C/O < 1) to recent AGB evolution and pulsation models (Marigo et al. 2017; Trabucchi et al. 2019, 2021). To this end, we show in Figure 4 the evolutionary tracks ¹⁴ of surface composition and pulsation periods for different initial masses.

Zoom In Zoom Out Reset image size

AGB stars with lower initial masses (≲1.5 M_⊙) do not evolve to the observed long period (Vassiliadis & Wood 1993; Trabucchi et al. 2019). While the 2.3 M_⊙ model does evolve to P > 500 days, stars in the initial-mass range ≈2.0–2.5 M_⊙ are known to evolve to carbon Miras, inconsistent with the O-rich composition. As shown in Figure 4, we find that stars in the initial-mass range ≈3.0–5.0 M_⊙ evolve to the long observed period with an O-rich composition (due to the hot bottom burning process destroying carbon in this higher-mass range; Höfner & Olofsson 2018). We thus favor the scenario where the donor in Sct X-1 had an initial mass of ≈3.0–5.0 M_⊙ star, corresponding to a current binary system age of ≈0.1–0.3 Gyr. Our conclusions are consistent with the period–age relationship of Mira variables discussed in the recent work of Trabucchi & Mowlavi (2022, their Figures 2 and B.1) and suggest that Sct X-1 is an intermediate-mass X-ray binary.

AGB donors have been long predicted to exist in X-ray binaries based on orbits of radio pulsars (e.g., van den Heuvel 1994); however, known radio pulsars probe binary populations with closer orbits (∼days to months). On the other hand, the formation of wide-orbit SyXRBs requires fine-tuned conditions. If the progenitor binary is too wide, it is likely to get disrupted during the preceding supernova (SN) or never reach the high wind accretion rate required to produce symbiotic activity. If the binary is too close, the NS may merge with its companion during the late-stage evolution of the initially lower-mass secondary star (Iben et al. 1995). The inferred temperature (≈2000 K) and luminosity (≈10⁴ L_⊙) suggest a Mira radius of ≈850 R_⊙. Taking the current Mira and NS mass to be ≈3 M_⊙ and ≈1.3 M_⊙, respectively, the inferred radius constrains the orbital period to be ≳12 yr if the accretion is wind fed (Kaplan et al. 2007), making Sct X-1 possibly the longest orbital period SyXRB known (Hinkle et al. 2019).

Population synthesis calculations (Lü et al. 2012; Yungelson et al. 2019) suggest that AGB donors should be rare among wind-accreting SyXRBs. The NS spin period in Sct X-1 overlaps with the predicted range for AGB donors (Figure 3 in Yungelson et al. 2019), but our estimated orbital period is larger than predicted in their models (≲10³ days). This discrepancy is a consequence of their models predicting wind-accreting SyXRBs only in the donor's early AGB phase in closer orbit binaries. Specifically, Lü et al. (2012) suggested that AGB donors in the thermally pulsing phase should overflow their Roche lobes, resulting in luminous X-ray emission (L_X ≳ 10³⁶ erg s⁻¹) unlike known wind-fed SyXRBs. This is also inconsistent with the recent low recent X-ray luminosity of Sct X-1 (L_X ≲ 10³⁵ erg s⁻¹) that is suggestive of wind-fed accretion (Kaplan et al. 2007) from a thermally pulsing AGB star and hence requires a wider orbit than produced in their calculations.

Despite being one of the brightest X-ray sources known from the early days of X-ray astronomy, Sct X-1 demonstrates the potential of wide-field infrared time-domain surveys combined with broadband infrared spectroscopy on moderate-aperture telescopes to understand this population of intrinsically red and obscured sources. We note that donor classifications in SyXRBs have been commonly obtained from optical spectroscopy, while the NIR region contains the most sensitive tracers of the spectral type and luminosity class for late-type evolved stars (Messineo et al. 2021). We thus advocate for a uniform NIR spectroscopic and photometric classification of SyXRBs to constrain their true demographics. In particular, the existence of a significant population of Mira AGBs in SyXRBs would suggest the need for a larger fraction of wide-orbit SyXRB progenitors as well as small birth kicks during the NS formation to keep these systems bound after the SN explosion. In fact, Yungelson et al. (2019) suggested that nearly 80% of SyXRBs may arise from electron-capture SNe due to the resulting low birth kicks. Alternatively, Hinkle et al. (2006) suggested that the evolution scenario in SyXRBs creates ideal conditions for the formation of NSs via the accretion-induced collapse of WDs, making them uniquely suited to search for this predicted population.

We thank S. Rappaport, M. Boyer, and D. Kaplan for valuable discussions. Support for this work was provided by NASA through the NASA Hubble Fellowship grant #HST-HF2-51477.001 awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5-26555. R.S. acknowledges grant number 12073029 from the National Science Foundation of China. We acknowledge the Visiting Astronomer Facility at the Infrared Telescope Facility, which is operated by the University of Hawaii under Cooperative Agreement no. NNX-08AE38A with the National Aeronautics and Space Administration, Science Mission Directorate, Planetary Astronomy Program. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Maunakea has always had within the indigenous Hawaiian community. PGIR is generously funded by Caltech, Australian National University, the Mt Cuba Foundation, the Heising-Simons Foundation, the Bi-national Science Foundation. PGIR is a collaborative project among Caltech, Australian National University, University of New South Wales, Columbia University, and the Weizmann Institute of Science. The scientific results reported in this article are based in part on data obtained from the Chandra Data Archive. This research has made use of software provided by the Chandra X-ray Center (CXC) in the application packages CIAO, ChIPS, and Sherpa.

Using follow-up observations using the MagIC instrument on the Magellan Baade telescope, Kaplan et al. (2007) reported 2MASS J1835 had optical fluxes of r > 25.2 mag and i ≈ 23.6 mag. However, the source was reported in the Gaia catalog at G ≈ 20.4 mag, G_RP ≈ 18.35 mag, and G_BP ≈ 22.34 mag. Here, we investigate this discrepancy accounting for the different filters used in the observations. Because Sct X-1 is undetected at λ < 1 μm and the IRTF library is limited to λ > 0.7 μm, we use the optical spectrum of the M6-III Mira variable V1334 Sgr from the X-shooter spectral library (Gonneau et al. 2020) to estimate the effects of the heavy reddening on the observed magnitudes. We simulate the extinction to Sct X-1 by reddening the observed spectrum of V1334 Sgr with E(B − V) = 6.5 (assuming a Cardelli et al. 1989 extinction law with R_V = 3.1), such that the resulting J − K color of the comparison star is the same as 2MASS J1835. Performing photometry on the reddened spectrum using the respective filter profiles, ¹⁵ we find expected colors of G_RP − r ≈ 8.5 mag, G_RP − i ≈ 2.7 mag, and G_BP − G_RP ≈ 10 mag.

Thus, the large difference between the G_RP and r filters for such red sources explains the nondetection of 2MASS J1835 in the r-band observation of Kaplan et al. (2007). However, the relatively blue observed color of G_BP − G_RP ≈ 4 mag compared to the simulated M6-III star suggests that pulsational variability is also required to explain the observations. Although the exact dates of observation are not available in Gaia EDR3, the EDR3 observations were obtained over a period of ≈34 months, spanning ≈1.5 cycles of the estimated pulsation period. For the observed J-band amplitude of ≈3.5 mag, the estimated amplitudes in optical filters would be ≳6 mag given the typical wavelength dependence of amplitude in O-rich Miras (Iwanek et al. 2021). Gaia EDR3 reports phot_bp_n_obs = 2 and phot_rp_n_obs = 9 for 2MASS J1835, suggesting that the source was indeed detected in more epochs in the redder filter where the source is brighter and has smaller flux amplitudes. On the other hand, the G_BP flux is likely dominated by observation epochs near the pulsation maximum where Miras are bluest in color (Reid & Goldston 2002). Thus, the reported Gaia EDR3 colors are expected to be closer to the bluest color of the source near maximum light.