NICER Discovery of Millisecond X-Ray Pulsations and an Ultracompact Orbit in IGR J17494-3030

Accreting millisecond X-ray pulsars (AMXPs; see Di Salvo & Sanna 2020, for a recent review) are rapidly rotating, weakly magnetized ( ∼ 10⁸ G) neutron stars accreting from a low-mass (≲1 M_⊙) companion in a low-mass X-ray binary (LMXB). Most known AMXPs are X-ray transient systems in which long (∼years) intervals of X-ray quiescence are punctuated by brief (∼weeks) outbursts of enhanced X-ray emission. These transient outbursts are understood to arise from a thermal instability in the accretion disk around a neutron star or black hole LMXB primary, analogous to "dwarf nova" optical outbursts in accreting white dwarfs (WDs; see Lasota 2001; Hameury 2020, and references therein).

The X-ray transient IGR J17494−3030 (Galactic coordinates l = 3591, b = −15; hereafter called IGR J17494) was first discovered in a 2012 March outburst in the 3–80 keV hard X-ray band (IBIS and JEM-X) in an INTEGRAL survey of the Galactic center region (Boissay et al. 2012). Soft X-ray (0.5–10 keV) monitoring observations with Swift showed that the outburst lasted approximately 1 month (Armas Padilla et al. 2013) before fading into quiescence (Chakrabarty et al. 2013). XMM-Newton 0.5–10 keV spectroscopy suggested that the compact primary is a neutron star (Armas Padilla et al. 2013). A new outburst was detected with INTEGRAL in 2020 October (Ducci et al. 2020), leading to a more precise X-ray localization with Chandra (Chakrabarty & Jonker 2020) and the identification of a 4.5 GHz radio counterpart with the Very Large Array (van den Eijnden et al. 2020).

Soft X-ray observations of the 2020 outburst with the Neutron Star Interior Composition Explorer (NICER) revealed the presence of coherent 376 Hz pulsations modulated by a 75-minute binary orbit, establishing the system as a millisecond pulsar (neutron star) in an ultracompact binary (Ng et al. 2020). In this letter, we first outline the NICER and XMM-Newton observations and data processing. We then present results from timing and spectral analyses of the NICER observations, as well as from a timing analysis of the archival 2012 XMM-Newton observations. Finally, we constrain the possible nature of the donor in the IGR J17494 system and discuss further implications of the source.

2.1. NICER

NICER is an X-ray telescope mounted on the International Space Station (ISS) since 2017 June. NICER has 56 aligned pairs of X-ray concentrator optics and silicon drift detectors (52 detectors are usually active on NICER). NICER is capable of fast-timing observations in the 0.2–12.0 keV band, with timing accuracy of time-tagged photons to better than 100 ns (Gendreau et al. 2012; LaMarr et al. 2016; Prigozhin et al. 2016).

NICER observed IGR J17494 from 2020 October 27 to November 4¹⁴ for a total exposure time of 32.3 ks after filtering, in ObsIDs 3201850101–3201850108.¹⁵ These observations were available through the public NASA HEASARC data archive. There were additional NICER observations, to which we did not have access, during this interval for a proprietary guest observer investigation (PI: A. Sanna; shown as the shaded region in the top panel of Figure 1). The events were barycenter corrected in the ICRS reference frame, with source coordinates R.A. = 267348417 and decl. = −30499722 (equinox J2000.0) obtained from a recent Chandra observation (Chakrabarty & Jonker 2020), using barycorr from FTOOLS with the JPL DE405 solar system ephemeris (Standish 1998).

Zoom In Zoom Out Reset image size

The NICER observations were processed with HEASoft version 6.28 and the NICER Data Analysis Software (nicerdas) version 7.0 (2020-04-23_V007a). The following criteria, which we note are relaxed compared to standard filtering criteria because the latter were too restrictive and resulted in no events, were imposed in the construction of the good time intervals (GTIs): no discrimination of events when NICER (on the ISS) was inside or outside of the South Atlantic Anomaly during the course of the observations; ≥20° for the source–Earth limb angle (≥30° for the Sun-illuminated Earth); ≥38 operational Focal Plane Modules (FPMs); undershoot (dark current) count-rate range of 0–400 per FPM (underonly_range); overshoot (saturation from charged particles) count-rate range of 0–2 per FPM (overonly_range and overonly_expr); pointing offset is <0015 from the nominal source position.

We analyzed spectral data using XSPEC v12.11.1 (Arnaud 1996). NICER data were selected in the range of 1–10 keV, to avoid contamination from optical loading and significant interstellar absorption at lower energy. The spectra were rebinned to have at least 25 counts per bin. Background spectra were extracted using nibackgen3C50 version 6 from the official NICER tools.¹⁶ Standard response files made available by the NICER team were used to perform spectral analysis.¹⁷

2.2. XMM-Newton

3.1. NICER

The NICER 1–7 keV light curve for the 2020 outburst is shown in the top panel of Figure 1. The source gradually faded until MJD 59155.4, after which it decayed more rapidly. The X-ray spectrum prior to the proprietary data gap was fairly constant and well fit with a two-component absorbed power-law and blackbody model (tbabs(powerlaw+bbodyrad)in XSPEC), with absorption column density n_H = 2.07(6) × 10²² cm⁻², photon index Γ = 1.90(6), blackbody temperature kT = 0.58(3) keV, and blackbody radius R_bb = 2.9(5)d₁₀ km, where d₁₀ is the source distance in units of 10 kpc. The uncertainties are reported at the 90% confidence level. The reduced χ² (${\chi }_{\nu }^{2}$) of the fit was 1.14 for 849 degrees of freedom. The spectrum softened during the late decay phase of the outburst, where the same two-component model fit yielded ${\rm{\Gamma }}={4.3}_{-0.6}^{+0.9}$ and n_H is assumed to be unchanged throughout the observations. The peak absorbed 1–10 keV flux we observed was 1.01 × 10⁻¹⁰ erg s⁻¹ cm⁻² on MJD 59149.4, corresponding to an unabsorbed flux of 1.43 × 10⁻¹⁰ erg s⁻¹ cm⁻². The lowest absorbed flux we measured was 1.21 × 10⁻¹² erg s⁻¹ cm⁻² on MJD 59157.5, corresponding to an unabsorbed flux of 3.23 × 10⁻¹² erg s⁻¹ cm⁻². This is roughly a factor of 3 fainter than the minimum flux detected by XMM-Newton at the end of the 2012 outburst (Armas Padilla et al. 2013). A more detailed X-ray spectral analysis will be reported elsewhere.

We first detected X-ray pulsations with a data analysis pipeline that employs multiple techniques¹⁹ for X-ray pulsation searches, including averaged power spectral stacking with Bartlett's method (Bartlett 1948) and acceleration searches with PRESTO (Ransom et al. 2002). The initial detection was made through PRESTO, an open-source pulsar timing software package²⁰ designed for efficient searches for binary millisecond pulsars. We ran a Fourier-domain acceleration search scheme with the accelsearch task over the range 1–1000 Hz and posited that the Doppler motion would cause the possible signal to drift over a maximum of 100 bins in Fourier frequency space. This yielded a strong ≃376.05 Hz pulsation candidate (trial-adjusted significance of 3.5σ) in the 2–12 keV range.

After initial identification of the candidate in the 2–12 keV range, we optimized the pulse significance by adjusting the energy range to maximize the ${Z}_{1}^{2}$ statistic, where

$$\begin{eqnarray}&&{Z}_{1}^{2}=\displaystyle \frac{2}{N}\left[{\left(\displaystyle \sum _{j=1}^{N}\cos 2\pi \nu {t}_{j}\right)}^{2}+{\left(\displaystyle \sum _{j=1}^{N}\sin 2\pi \nu {t}_{j}\right)}^{2}\right],\end{eqnarray} \tag{ 1 }$$

where t_j are the N photon arrival times (Buccheri et al. 1983). We found that an optimal energy range of 1.01–7.11 keV yielded ${Z}_{1}^{2}=1915.41$. Our subsequent timing analyses were carried out over 1–7 keV.

The acceleration searches indicated that the pulsation frequency is modulated by a binary orbit. We used the acceleration data to estimate an initial timing model with a provisional circular orbit. We then used this initial model to construct 35 pulse times of arrival (TOAs) with the photon_toa.py tool in the NICERsoft²¹ data analysis package, using a Gaussian pulse template and ensuring an integration time of 500 s for each TOA (with minimum exposure time of 200 s). We then used these TOAs to compute corrections to our initial orbit model using weighted least-squares fitting with the PINT pulsar data analysis package (Luo et al. 2020). Our best-fit orbit ephemeris is shown in Table 1, and the orbital decay curve is shown in the middle panel of Figure 1. Using our best-fit timing model, pulsations were detected throughout the entire outburst. At the end of the observations, we were able to detect the pulsations in observations from MJD 59154–59157 (November 1–4) by combining all the data. The mean unabsorbed flux over this 4-day interval was 8.5 × 10⁻¹² erg s⁻¹ cm⁻² (1–10 keV). We did not have sufficient sensitivity to detect the pulsations in individual pointings from these dates. The time-averaged fractional rms pulsed amplitude was 7.4% (1–7 keV). Examining the lower and higher energies separately, we found amplitudes of 7.2% in the 1–3 keV band and 8.7% in the 3–7 keV band. The soft and hard X-ray pulse profiles are shown in the bottom panel of Figure 1. The 1–3 keV profile shows the presence of a second harmonic; this component is not significantly detected in the 3–7 keV profile. To further examine the energy dependence of the pulse waveform, we adaptively binned the timing data in energy. We required the energy bins to contain a multiple of five pulse-invariant (PI) energy channels (0.05 keV), such that each bin contained at least 5000 counts. For each of these energy bins, we then folded the data using our best-fit timing solution and measured the background-corrected fractional rms pulsed amplitude and the pulse phase offset relative to the model. The resulting energy dependencies are shown in Figure 2. The pulsed amplitude has a local maximum of 11% at 4 keV, while the pulse phase lag has a local maximum of +0.05 cycles (130 μs) at around 1.5 keV.

Zoom In Zoom Out Reset image size

Table 1.  IGR J17494−3030 Timing Parameters from the 2020 Outburst

Parameter	Value
R.A., α (J2000)	267348417
Decl., δ (J2000)	−30499722
Position epoch (TT)	MJD 59156.34
Spin frequency, ν0 (Hz)	376.05017022(4)
Spin frequency derivative (during outburst), $| \dot{\nu }|$ (Hz s−1)	<1.8 × 10−12
Spin epoch, t0 (TDB)	MJD 59149.0
Binary period, Porb (s)	4496.67(3)
Projected semimajor axis, ${a}_{x}\sin i$ (lt-ms)	15.186(12)
Epoch of ascending node passage, Tasc (TDB)	MJD 59149.069012(15)
Eccentricity, e	<0.006 (2σ)
Spin frequency derivative (long-term), $\dot{\nu }$ (Hz s−1)	− 2.1(7) × 10−14

Note. Source coordinates adopted for the barycentering were determined by Chakrabarty & Jonker (2020). The spin frequency derivative quoted here is during the 2020 outburst.

Download table as:  ASCIITypeset image

3.2. XMM-Newton

The discovery of coherent millisecond X-ray pulsations from IGR J17494−3030 definitively identifies the source as an accreting neutron star. We can use the long-term spin-down of the pulsar between its 2012 and 2020 X-ray outbursts to estimate the pulsar's magnetic field strength. Assuming that the spin-down is due to magnetic dipole radiation, we can calculate the pulsar's magnetic dipole moment (Spitkovsky 2006)

$$\begin{eqnarray}\begin{array}{rcl}\mu & = & 5.2\times {10}^{26}{\left(1+{\sin }^{2}\alpha \right)}^{-1/2}\\ & & \times {\left(\displaystyle \frac{I}{{10}^{45}{\rm{g}}{\mathrm{cm}}^{2}}\right)}^{1/2}\,{\rm{G}}\ {\mathrm{cm}}^{3},\end{array}\end{eqnarray} \tag{ 2 }$$

where α is the angle between the magnetic and spin axes and I is the neutron star moment of inertia. The corresponding surface dipole field strength of ≃10⁹ G is on the high end of the distribution inferred for other AMXPs (Mukherjee et al. 2015).

We found that the fractional rms pulsed amplitude and the pulse phase of IGR J17494 vary as a function of photon energy. Both the amplitude and the phase lag reach a local maximum at a (different) characteristic energy of 4 and 1.5 keV, respectively. Energy-dependent variations of the pulse waveform are ubiquitous among AMXPs, although the location of these local maxima varies greatly from source to source (Gierliński et al. 2002; Falanga et al. 2005, 2012; Gierliński & Poutanen 2005; Patruno et al. 2009). The behavior can be understood through a two-component emission model, with thermal emission originating from the stellar surface and scattered Compton emission originating from some height above the surface (Gierliński et al. 2002; Wilkinson et al. 2011). Accounting for the difference in geometry and emission patterns, such a model can self-consistently explain the energy dependence of both the phase lags and the pulsed amplitudes (Poutanen & Gierliński 2003).

Our measurement of a 75-minute binary orbit allows us to constrain the nature of the mass donor in this system. The vast majority of Roche-lobe-filling LMXBs and cataclysmic variables contain hydrogen-rich donor stars, and they all have binary periods P_orb ≳ 80 minutes (Paczynski & Sienkiewicz 1981; Rappaport et al. 1982). The so-called ultracompact binaries (P_orb ≲ 80 minutes) have H-depleted donors (Nelson et al. 1986; Pylyser & Savonije 1988, 1989; Nelemans et al. 2010). IGR J17494 has the longest known period for an ultracompact LMXB and lies near the period boundary, making it a particularly interesting case. We also note the recent discovery of the rotation-powered millisecond gamma-ray pulsar PSR J1653−0158 in a 75-minute (nonaccreting) binary (Nieder et al. 2020). This is the shortest orbital period known for a rotation-powered binary pulsar, and this "black widow" system is believed to have evolved from an ultracompact LMXB after mass transfer ended.

From our measured orbital parameters, the binary mass function of IGR J17494 is

$$\begin{eqnarray}\begin{array}{rcl}{f}_{m} & \equiv & \displaystyle \frac{{\left({M}_{d}\sin i\right)}^{3}}{{\left({M}_{\mathrm{ns}}+{M}_{d}\right)}^{2}}=\displaystyle \frac{4{\pi }^{2}{\left({a}_{x}\sin i\right)}^{3}}{{{GP}}_{\mathrm{orb}}^{2}}\\ & & \approx 1.39\times {10}^{-6}\,{M}_{\odot },\end{array}\end{eqnarray} \tag{ 3 }$$

where M_ns is the neutron star mass, M_d is the donor mass, ${a}_{x}\sin i$ is the projected semimajor axis, and the binary inclination i is defined as the angle between the line of sight and the orbital angular momentum vector. For a given value of M_ns, we can use Equation (3) to calculate M_d as a function of i (see top panel of Figure 3). Assuming M_ns = 1.4 (2.0) M_⊙, the minimum donor mass (for an edge-on binary with i = 90°) is 0.014 (0.018) M_⊙. For a random ensemble of binaries, the probability distribution of $\cos i$ is uniformly distributed and $\Pr (i\lt {i}_{0})=1-\cos {i}_{0}$. Thus, the donor mass is likely to be very low, with a 90% confidence upper limit of M_d < 0.033 (0.041) M_⊙ for M_ns = 1.4 (2.0) M_⊙.

Zoom In Zoom Out Reset image size

Assuming a Roche-lobe-filling donor, we can calculate the donor radius R_d as a function of M_d (Eggleton 1983); this is shown in the bottom panel of Figure 3 for M_ns = 1.4 M_⊙. For comparison, the figure also shows the mass–radius relations for different types of low-mass stars: cold WDs (Zapolsky & Salpeter 1969; Rappaport & Joss 1984; Nelemans et al. 2001); hot (finite-entropy) WDs composed of either He, C, or O (Deloye & Bildsten 2003); and low-mass H-rich stars, including brown dwarfs (Chabrier et al. 2000). We see that cold WD models are inconsistent with our measured mass–radius constraint, indicating that thermal bloating is likely important. Moderately hot He WDs with central temperature T_c = 2.5 × 10⁶ K or C/O WDs with T_c = 5 × 10⁶ K are consistent with our constraint at high binary inclination. Hotter WDs and moderately old (cool) brown dwarfs are also consistent, but the required inclinations have low a priori probability. Finally, H-rich dwarfs above the mass-burning limit are also possible, but only for extremely low (improbable) inclinations. We conclude that the donor is likely to be a ≃0.02 M_⊙ finite-entropy He or C/O WD.

The angular momentum evolution of the binary is described by (Verbunt 1993; Verbunt & van den Heuvel 1995)

$$\begin{eqnarray}&&-\displaystyle \frac{\dot{J}}{J}=-\displaystyle \frac{{\dot{M}}_{d}}{{M}_{d}}\,{f}_{\mathrm{ML}},\end{eqnarray} \tag{ 4 }$$

where $\dot{J}$ is the rate of change of the orbital angular momentum J due to effects other than mass loss from the system, ${\dot{M}}_{d}$ (<0) is the rate of change of the donor mass, and the dimensionless factor f_ML is given by

$$\begin{eqnarray}&&{f}_{\mathrm{ML}}=\displaystyle \frac{5}{6}+\displaystyle \frac{n}{2}-\beta q-\displaystyle \frac{(1-\beta )(q+3\alpha )}{3(1+q)},\end{eqnarray} \tag{ 5 }$$

where q = M_d/M_ns ≪ 1 is the binary mass ratio, β is the fraction of ${\dot{M}}_{d}$ that accretes onto the neutron star (β = 1 for conservative mass transfer),

$$\begin{eqnarray}&&n=\displaystyle \frac{d(\mathrm{ln}{R}_{d})}{d(\mathrm{ln}{M}_{d})}\end{eqnarray} \tag{ 6 }$$

denotes how the donor radius R_d changes with mass loss, and α is the specific angular momentum of any (nonconservative) mass lost from the system in units of the donor star's specific angular momentum. Thus, α parameterizes the site of any mass ejection from the system, where α ≃ 1 for mass loss close to the donor and α ≃ q² for mass loss close to the pulsar. Mass transfer in ultracompact binaries is primarily driven by angular momentum loss due to gravitational radiation from the binary orbit (see Rappaport et al. 1982, and references therein); for a circular orbit, this loss is given by (Landau & Lifshitz 1989; Peters 1964)

$$\begin{eqnarray}&&-{\left(\displaystyle \frac{\dot{J}}{J}\right)}_{\mathrm{GW}}=\displaystyle \frac{32\,{G}^{3}}{5\,{c}^{5}}\displaystyle \frac{{M}_{\mathrm{ns}}{M}_{d}({M}_{\mathrm{ns}}+{M}_{d})}{{a}^{4}},\end{eqnarray} \tag{ 7 }$$

where a is the binary separation. Inserting this into the left-hand side of Equation (4), we can then calculate the gravitational-wave-driven mass transfer rate from the donor into the accretion disk as

$$\begin{eqnarray}\begin{array}{rcl}{\dot{M}}_{\mathrm{GW}} & = & -{\dot{M}}_{d}=\displaystyle \frac{32{G}^{3}}{5{c}^{5}}{\left(\displaystyle \frac{4{\pi }^{2}}{G}\right)}^{4/3}\displaystyle \frac{{M}_{\mathrm{ns}}^{8/3}\,{q}^{2}}{{\left(1+q\right)}^{1/3}\,{P}_{\mathrm{orb}}^{8/3}\,{f}_{\mathrm{ML}}}\\ & \approx & 2.6\times {10}^{-12}{\left(\displaystyle \frac{{M}_{\mathrm{ns}}}{1.4{M}_{\odot }}\right)}^{2/3}\\ & & \times {\left(\displaystyle \frac{{M}_{d}}{0.014{M}_{\odot }}\right)}^{2}{\left(\displaystyle \frac{{f}_{\mathrm{ML}}}{0.66}\right)}^{-1}\,{M}_{\odot }\,{\mathrm{yr}}^{-1}.\end{array}\end{eqnarray} \tag{ 8 }$$

Our scaling value of f_ML = 0.66 corresponds to n = −1/3 (typical for degenerate donors) and β = 1.

Although accretion onto the neutron star is mediated by episodic outbursts, mass continuity requires that the long-term average accretion luminosity reflect ${\dot{M}}_{\mathrm{GW}}$ if the mass transfer is conservative. Our observations are not ideal for examining this, since we did not observe the early (brightest) part of the 2020 outburst with NICER. However, the unabsorbed 0.5–10 keV X-ray fluence in the 2012 outburst was 1.1 × 10⁻⁴ erg cm⁻² (Armas Padilla et al. 2013). Assuming that the 2012 outburst was typical, that the long-term average accretion rate is dominated by the outbursts, and that there were no intervening outbursts between 2012 and 2020, the outburst separation of ≈3100 days yields a long-term average X-ray flux of F_x,avg = 3.9 × 10⁻¹³ erg s⁻¹ cm⁻² (0.5–10 keV). We can then write the accretion luminosity as

$$\begin{eqnarray}&&\displaystyle \frac{{{GM}}_{\mathrm{ns}}\beta {\dot{M}}_{\mathrm{GW}}}{{R}_{\mathrm{ns}}}=\left(\displaystyle \frac{{\rm{\Delta }}{\rm{\Omega }}}{4\pi }\right)4\pi {d}^{2}{f}_{\mathrm{bol}}\,{F}_{x,\mathrm{avg}},\end{eqnarray} \tag{ 9 }$$

where R_ns is the neutron star radius, d is the distance to the source, f_bol is the bolometric correction (accounting for accretion luminosity outside the 0.5–10 keV bandpass), and ΔΩ is the solid angle into which the accretion luminosity is emitted. Based on the INTEGRAL hard X-ray observations in 2012 (Boissay et al. 2012), we estimate f_bol ≈ 1.7. Assuming R_ns = 10 km and taking β = 1 and ΔΩ = 4π, we obtain an implausibly large distance of 20 kpc. Although it is not impossible that the source lies on the far side of the Galaxy, a location near the Galactic center is far more likely given the line of sight. There are several reasons that our distance estimate might be significantly inflated. Obtaining a more plausible distance of 8 kpc would require

$$\begin{eqnarray}&&\begin{array}{l}\displaystyle \frac{1}{\beta }\left(\displaystyle \frac{{\rm{\Delta }}{\rm{\Omega }}}{4\pi }\right)\left(\displaystyle \frac{{f}_{\mathrm{bol}}}{1.7}\right)\left(\displaystyle \frac{{f}_{\mathrm{ML}}}{0.66}\right)\\ \,\times \,{\left(\displaystyle \frac{{M}_{\mathrm{ns}}}{1.4{M}_{\odot }}\right)}^{-5/3}{\left(\displaystyle \frac{{M}_{d}}{0.014{M}_{\odot }}\right)}^{-2}\\ \,\times \,\left(\displaystyle \frac{{F}_{x,\mathrm{avg}}}{3.9\times {10}^{-13}\,\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}}\right)\approx 6.\end{array}\end{eqnarray} \tag{ 10 }$$

Some combination of these factors may be different than what we assumed above. However, a heavier neutron star (M_ns > 1.4 M_⊙), a heavier mass donor (equivalent to a lower binary inclination), or significant beaming (ΔΩ < 4π) would further inflate the distance estimate. Also, our estimate of f_bol is fairly robust, given the broad X-ray coverage of the INTEGRAL data. It is possible that we have underestimated F_x,avg. This could happen if we missed accretion outbursts that occurred between 2012 and 2020, or if the quiescent (nonoutburst) flux is as high as ∼10⁻¹² erg s⁻¹ cm⁻². The former possibility can be explored through a careful analysis of archival X-ray monitoring data, while the latter possibility could be checked through sensitive X-ray observations of the source in quiescence.

The factor f_ML may be somewhat larger than we assumed. Although we calculated it using the usual value of n = −1/3 for degenerate donors, Deloye & Bildsten (2003) showed that the WD donors in ultracompact binaries can have n values in the range of −0.1 to −0.2 owing to the importance of Coulomb interactions for extremely low donor masses. However, this is unlikely to increase f_ML by more than a factor of ≃1.2.

Nonconservative mass transfer (β < 1) is a more promising avenue. The radio detection of IGR J17494 (van den Eijnden et al. 2020) points to the likelihood of a collimated jet ejection during the outburst. Moreover, a similar distance conundrum was invoked to infer nonconservative mass transfer in the ultracompact LMXB pulsar XTE J0929−314 (Marino et al. 2017), as well as several other AMXPs (Marino et al. 2019). Also, there was evidence found for an outflow in the ultracompact LMXB pulsar IGR J17062−6143 (Degenaar et al. 2017; van den Eijnden et al. 2018), possibly arising from a magnetic propeller-driven wind from the inner accretion disk (Illarionov & Sunyaev 1975).

During the long periods of X-ray (accretion) quiescence, mass loss from the binary could arise from several different mechanisms. These are motivated by the study of rotation-powered radio millisecond pulsars in detached (nonaccreting) binaries: the so-called "black widow" (M_c ≲ 0.05 M_⊙) and "redback" (M_c ≳ 0.1 M_⊙) systems, where M_c is the companion mass (see, e.g., Romani et al. 2016, and references therein). One possibility is black-widow-like ablation of the companion, driven by rotation-powered gamma-ray emission from the pulsar (Ginzburg & Quataert 2020). Such ablation could also be driven by particle heating via the rotation-powered pulsar wind (see Harding & Gaisser 1990, and references therein). Hard X-rays and gamma rays from the intrabinary shock observed in many black widow systems could significantly affect the mass-loss rate (Wadiasingh et al. 2018). Another possibility is that the pulsar wind could drive an outflow from the inner Lagrange (L₁) point by overcoming the ram pressure of accreting material (Burderi et al. 2001; Di Salvo et al. 2008).

As an example, we consider the case of gamma-ray ablation. If we assume that the gamma-ray luminosity is ≃10% of the spin-down luminosity (≃3 × 10³⁵ erg s ⁻¹ based on our long-term $\dot{\nu }$ measurement) as typically seen in black widow systems (Abdo et al. 2013), this would imply a companion mass-loss rate of ∼10⁻¹¹M_⊙ yr⁻¹ (Ginzburg & Quataert 2020). For a source distance of 8 kpc and assuming that gravitational wave losses dominate in Equation (4), this implies β ≈ 0.04 and α ≈ 0.4, suggesting that the mass ejection occurs somewhere between the pulsar and the L₁ point (α ≈ 0.8). However, Ginzburg & Quataert (2020) argue that magnetic braking of the donor (through magnetic coupling to the ablated wind) likely dominates gravitational radiation as an angular momentum sink in black widow systems. If so, then that could both decrease β and increase α even further in our case.

All of the X-ray–quiescent mechanisms mentioned above rely on the system entering a rotation-powered radio pulsar state during X-ray quiescence. We note that a growing class of so-called transitional millisecond pulsars (tMSPs) has been identified that switch between LMXB and radio pulsar states (see Papitto & de Martino 2020, for a review). The known tMSPs would be classified as redback systems in their radio pulsar state. If IGR J17494 is a tMSP, then its low companion mass would make it a black widow system in its rotation-powered state. We note that the X-ray properties of IGR J17494 correspond to those of the so-called very faint X-ray transients (VFXTs; Wijnands 2006), whose low outburst luminosities and long-term accretion rates are difficult to understand. Our observations support the suggestion that some VFXTs may also be tMSPs (Heinke et al. 2015). The distinction between VFXTs and ordinary LMXBs may somehow relate to the level of nonconservative mass transfer.

M.N. and this work were supported by NASA under grant 80NSSC19K1287, as well as through the NICER mission and the Astrophysics Explorers Program. NICER work at NRL is also supported by NASA. D.A. acknowledges support from the Royal Society. This research has made use of data and/or software provided by the High Energy Astrophysics Science Archive Research Center (HEASARC), which is a service of the Astrophysics Science Division at NASA/GSFC and the High Energy Astrophysics Division of the Smithsonian Astrophysical Observatory.

Facilities: NICER - , XMM. -

Software: astropy (Astropy Collaboration et al. 2013), NumPy and SciPy (Oliphant 2007), Matplotlib (Hunter 2007), IPython (Perez & Granger 2007), tqdm (Da Costa-Luis et al. 2020), NICERsoft, PRESTO (Ransom et al. 2002), PINT (Luo et al. 2020), HEASoft 6.28.