MAXI and NuSTAR Observations of the Faint X-Ray Transient MAXI J1848-015 in the GLIMPSE-C01 Cluster

Compact objects, such as neutron stars (NSs) and black holes (BHs), orbiting main-sequence stars often undergo cycles of outburst and quiescence due to modulation in the rate at which matter from the companion star accretes onto the compact object. Matter falls from the surface of the companion into the orbit of the compact object via Roche-lobe overflow or stellar winds (also known as Bondi-Hoyle accretion). According to the disk instability model (Lasota 2001), this material eventually reaches a critical density at which angular momentum can be transported efficiently outward, resulting in the formation of an accretion disk (Shakura & Sunyaev 1973). This disk may reach all the way down to the surface or innermost stable circular orbit (ISCO) of the compact object. As the gravitational potential energy of the accreting material is converted into heat, the innermost regions of the disk can reach temperatures of the order of 10⁷ K, emitting photons with energy exceeding thousands of electron volts; hence, these systems are referred to as X-ray binaries. This cycle of transient disk accretion leads to a variety of observable phenomena. The X-ray spectra of accreting NSs and BHs vary between a low-luminosity hard state and a brighter soft state, with intermediate states in between. During the hard state, emission is dominated by a power-law-shaped component that originates from a region of hot plasma near the central accretor known as the corona, while the soft state spectrum is dominated by thermal emission from the inner regions of the disk.

Additional spectral features result from the disk geometry in an X-ray binary. Iron in the disk may be irradiated by coronal emission and fluoresce, giving rise to emission lines around 6.4 keV. Line emission originating from the inner regions of the disk may be blurred by Doppler shifts due to the rapid orbital motion of the disk material and by the strong gravitational potential near the central compact object. Furthermore, soft photons from the corona may undergo Compton upscattering resulting in a "Compton hump" in the spectrum at high energies. These "reflection" features encode information regarding the inner disk radius, R_in, and the spin parameter, a, of the central accretor. Spectral models such as relxill (Dauser et al. 2014; García et al. 2014), which self-consistently model relativistic disk reflection, are therefore important tools for probing the properties of X-ray binaries, and can help us to differentiate between NS and BH accretors.

Each year, the Monitor of All-sky X-ray Image (MAXI; Matsuoka et al. 2009) discovers dozens of X-ray sources. Among these sources are accreting BHs, NSs, and white dwarfs. Follow-up with other X-ray observatories can help to elucidate the nature of these sources, and it lays the groundwork for their future study.

One such source, MAXI J1848015, was discovered with the MAXI/Gas Slit Camera (GSC) on 2020 December 20 (Takagi et al. 2020). The MAXI/GSC nova alert system (Negoro et al. 2016) triggered on the source at 05:04 (all times are given in UT), and the source flux was found to be increasing. The average X-ray flux on eight scan transits from 00:25 to 11:16 on December 20 was 63 ± 10 mCrab in the 4–10 keV band. The MAXI 90% confidence region had a radius of about 03 and was consistent with the previously detected ASCA source AX J1848.8-0129 (Sugizaki et al. 2001). Since the source had an angular separation of 26° from the Sun at the time of detection, neither the Neil Gehrels Swift Observatory X-ray Telescope (Swift/XRT; Gehrels et al. 2004; Burrows et al. 2005) nor the Neutron Interior Composition Explorer Mission (NICER; Gendreau & Arzoumanian 2017) could observe the source.

However, the Nuclear Spectroscopic Telescope Array (NuSTAR; Harrison et al. 2013) was able to perform follow-up observations of the source despite its angular proximity to the Sun. NuSTAR performed tiling observations to search the MAXI error region, and the source was detected during the first pointing. The source exhibited a soft spectrum, and although the source position was further refined to a region with radius ∼ 90'', it could still not be distinguished from AX J1848.8-0129 with certainty (Pike et al. 2020). Soon after the NuSTAR ToO observations, on December 23, the 2–10 keV source flux rapidly decreased in a day from about 40 mCrab to less than 15 mCrab (Negoro et al. 2020). About a week later, the source was again observed by NuSTAR, this time exhibiting a much harder spectrum (Mihara et al. 2021).

By the end of 2021 February, MAXI J1848015 was far enough from the Sun that it could be observed by other instruments. Thereafter, the source was localized by both Swift/XRT and the Chandra X-ray Observatory (Chandra; Weisskopf et al. 2000) to 90% confidence regions of ± 23 (Kennea et al. 2021) and ± 08 (Chakrabarty et al. 2021), respectively, while radio observations further refined the source position to α(J2000) = 18^h48^m49824 ± 0003, $\delta ({\rm{J}}2000)=-01^\circ 29^{\prime} 49\buildrel{\prime\prime}\over{.} 99\pm 0\buildrel{\prime\prime}\over{.} 05$ (Tremou et al. 2021). ¹⁰ Interestingly, the radio position is coincident with a relatively bright near-IR (NIR) counterpart (Hare et al. 2021). Around the same time, NICER observations of the source were performed, and Miller et al. (2021) reported a number of emission lines in the Fe K band and a flux of about 1 mCrab.

Importantly, the precise localization of the source confirmed that MAXI J1848015 is not consistent with the previously reported position of AX J1848.8-0129, and that it is spatially coincident with, and likely resides in, the core of the Galactic cluster GLIMPSE-C01 (GC01 hereafter). This cluster was originally discovered by Kobulnicky et al. (2005), who suggested that the cluster was an old globular cluster passing through the Galactic disk at a distance of 3.1–5.2 kpc. Several subsequent studies have alternatively suggested that GC01 is a young or intermediate-age massive cluster candidate with an age between 0.3–2.5 Gyr (Davies et al. 2011; Davidge et al. 2016). More recently, Hare et al. (2018) reported on Hubble Space Telescope observations of GC01, which they used to estimate a cluster distance of ∼3.3 kpc and a cluster age of >2 Gyr by studying the absolute magnitudes of red clump stars in the cluster. Unfortunately, the cluster's large source density, strong differential reddening across the cluster (ranging between A_V = 14 and 22), and unknown metallicity make it difficult to more precisely constrain the cluster's age (Hare et al. 2018).

In this paper, we present the results of MAXI monitoring of MAXI J1848015 as well as spectral and timing analysis of the two NuSTAR observations performed following the detection of the source by MAXI. In Section 2, we describe the MAXI observations in detail, including the shape and duration of the outburst. In Section 3, we discuss the NuSTAR observations of the source, beginning with a description of the observations and data reduction and continuing onto an investigation of the source spectra of, as well as the results of, reflection modeling applied to these spectra in Section 3.1. Next, we present an analysis of the timing properties of the NuSTAR light curves in Section 3.2. Finally, in Section 4, we discuss what our results mean in regards to the questions of whether the source is an NS or a BH accretor and how its particularly low luminosity can be understood in the context of disk accretion.

MAXI has been monitoring about 85% of sky every 92 minutes with the GSCs in the 2–20 keV band since 2009 August (Mihara et al. 2011; Sugizaki et al. 2011). The GSCs have two wide fields of view of 15 × 160° to the horizontal and zenith directions, and typically observe a source for 40–100 s every 92 minutes as the International Space Station (ISS) orbits Earth.

In 2020 December, the GSC_1, GSC_2, GSC_4, GSC_5, and GSC_7 cameras and the degraded GSC_3 and GSC_6 cameras were operating. The source was detected by all of these detectors, but in these analyses we only used those obtained with the well-calibrated GSC_4 and GSC_5 cameras with a high voltage of 1650V and GSC_2 and GSC_7 cameras with 1550V.

We employed a point-spread function (PSF) fit method to obtain MAXI/GSC light curves with the best signal-to-noise ratios (Morii et al. 2010). The count rate in each light-curve bin was obtained by fitting an image with the PSFs of the GSCs taking into account the presence of nearby sources such as the high-mass X-ray binary AX J1846.4-0258 (160 separation) and the super-giant fast X-ray transient IGR J18483−0311 (168 separation). On the other hand, the nearby transient sources in quiescent states, including Swift J185003.2−005627 (064), GS 1843−02 (093), and Swift J1845.7−0037 (111), were ignored. AX J1848.8−0129 (0012 ≃ 074) was also excluded. We note that AX J1848.8−0129 is originally named as AX J184848−0129 (Sugizaki et al. 2001), and its position has about 1' uncertainty. We first produced 1-scan, 6 hr, and 1 day bin light curves in the 2–4 keV and 4–10 keV bands, respectively. The 6 hr and 1 day bin data were obtained by fitting 4-scan and 16-scan image data, respectively. We then subtracted constant background components, mainly originating from the galactic ridge X-ray emission, as the average count rates from MJD 58,000 (2017 September 4) to 58599 (2019 April 26) were consistent with zero. Finally, we obtained 2–10 keV curves by summing the background-subtracted 2–4 keV and 4–10 keV curves.

As shown later, every ∼92 days (the ISS precession period), when the source was at the right side of the GSC_2 and GSC_5, the source was obscured by the Space-X Crew-1 spacecraft attached on the Harmony module of the ISS for several days. We did not use data during the periods shown in the current calibration database (CALDB), and we also excluded data from 1 day before and after this period due to some ambiguity regarding the shape of the spacecraft shadow.

Figure 1 shows the MAXI/GSC 2–4 keV and 4–10 keV light curves and their ratio for MAXI J1848015 obtained by the PSF-fit method. In the seven scan transits from 18:37 on December 18 to 01:12 on December 19 (MJD 59,201.7763–59,202.0501), the source was not visible on GSC images (Takagi et al. 2020). No count excess is recognized in each scan curve. The PSF-fit method provides two-sided 1σ errors based on the likelihood method (Morii et al. 2010). Fitting eight one-scan data points during the period with a constant model gives the average count rate of −0.094 ± 0.0128 ct cm⁻² s⁻¹. Using the size of this error, we estimate a 1σ upper limit of 0.0128 ct cm⁻² s⁻¹, ∼5.7 mCrab. ¹¹ This is consistent with a value 12 mCrab of a typical 3σ detection limit of 1 day of ∼16 scans (Negoro et al. 2016). After the scan transit at 01:12 until 22:52 on December 19 (MJD 59,202.05–59,202.95), the GSC did not fully observe the source region due to Sun avoidance. During the three scan transits from 11:59 to 15:05, however, some count excess was recognized at the edge of the detectors, which suggests the outburst started between 02:22 and 11:59.

Zoom In Zoom Out Reset image size

The light curves and hardness ratios in Figure 1 show that the 4–10 keV flux reached the peak in almost one day on December 20 (MJD 59,203) followed by gradual spectral softening. On December 23, the 2–10 keV flux rapidly decreased to below the detection limit in 1 day (Negoro et al. 2020) from 0.117 ± 0.016 ct cm⁻² s⁻¹ at MJD 59,205.89 (the center time of the 6 hr bin) to −0.003 ± 0.009 ct cm⁻² s⁻¹ at 59206.86 (we note that background subtraction may result in nonphysical negative count rates). However, we note that the light curve obtained during the first NuSTAR observation performed on December 23 indicates a flux increase during the 4 hr observation (see Section 3). This suggests that the rapid decrease observed by MAXI on December 23 is not simply due to the occultation by the companion. This flux drop continued until around December 26.0 (MJD 59,209.0), and the 4–10 keV flux increased again, possibly prior to an increase in the 2–4 keV flux.

We plot long-term variations in the flux of MAXI J1848015 in Figure 2. The data when the source is hidden by the Space-X Crew-1 spacecraft are not used and shown in gray. GIS images for the duration show extended enhancement around the source region. The origin of the enhancement is unknown, but it is difficult to consider that it is intrinsic to the source because of the periodicity closely connected with the interference and the extended image even though there is still some ambiguity in the shadow shape of the spacecraft.

Zoom In Zoom Out Reset image size

We also show the 2–10 keV light curve with a logarithmic time axis from the time December 22 04:48 (MJD 59,205.2 ≡ T) in the inset panel in Figure 2. We fit the decay with three different models: a linear decay, an exponential decay, and a power-law decay. The linear fit to the putative rapid decay starting at t₀( ≡ t − T) = 0.69 days (=MJD 59,205.89) is shown by the orange solid line. After t₀ = 3.92 days, the flux exhibited an exponential (blue curve) or a power-law (green) decay. Fitting the data for t₀ ≥ 3.92 days with an exponential function gives a time constant of 28.0 ± 9.4 days (shown by the blue solid line and its extrapolation by the dotted line). Fitting the flux decrease from t₀ ≥ 0.69 days to a power-law function, we obtained a power-law index of −0.97 ± 0.08, shown by the green solid and dotted line; however, we note that the result of the power-law fit depends strongly on the choice of T.

NuSTAR revealed a stark spectral change between the first and second observations. The spectra are shown in Figure 5 along with their best-fit models. We found that the spectrum during NuObs 1 was significantly softer than that observed during NuObs 2. We therefore refer to the former as the "soft state" and the latter as the "hard state." The spectra are described remarkably well by models that are frequently used to model the soft and hard states of accreting BHs and NSs: the continuum emission during the soft state is described well by an absorbed disk blackbody (diskbb) and a power law with index Γ, while the continuum emission during the hard state is described well by an absorbed power law with a high-energy cutoff, E_cut, which we modeled using cutoffpl. We modeled absorption using the tbabs model. We used cross sections provided by Verner et al. (1996) and abundances provided by Wilms et al. (2000).

Zoom In Zoom Out Reset image size

Each of the spectra exhibits broad emission lines around 6.4 keV corresponding to Fe Kα. Figure 6 shows the Fe line profiles that result from fitting the continuum emission while ignoring data bins between 5 and 8 keV. The broad, asymmetrical structure of the Fe line profiles led us to add relativistic disk reflection to the continuum models described above. We added a reflected component using relxilllp (Dauser et al. 2014; García et al. 2014) in order to self-consistently model emission originating from a hot corona with a lamppost geometry, which irradiates and is reprocessed by a thin disk. We tied the power-law index and high-energy cutoff ¹² of each power-law component to those of relxilllp. We found that this model was able to describe the spectrum well, achieving a fit statistic of W/d. o. f. = 449/323 and W/d. o. f. = 420/373 (and a reduced Chi-squared test statistic of ${\chi }_{\nu }^{2}=0.98$ and ${\chi }_{\nu }^{2}=1.01$) for the soft and hard states, respectively. However, the Fe line emission could not be fully accounted for, and residuals indicated the presence of a narrow line component. We therefore added a Gaussian component near 6.4 keV, and we froze its width at σ = 10⁻⁵ keV, much narrower than the instrumental energy resolution. We allowed the strength and the centroid of the line to vary while fitting, and we found that the addition of this narrow line improved the fit by ΔW = −17.5 and ΔW = −10.6 for the soft and hard states, respectively, while decreasing the number of degrees of freedom by 2. We also found that the addition of this component led to better constraints on other parameters, such as the spin and the inclination, while still remaining consistent with their previous estimates.

Zoom In Zoom Out Reset image size

In order to determine the statistical significance of the narrow components, we simulated 5000 spectra each for the soft and hard states, originating from the reflection models without narrow components. We then added a narrow component, again leaving the centroid energy and line strength free and fixing the width at σ = 10⁻⁵ keV. By calculating the resulting improvement in the fit statistic for each simulation, we arrived at a false-positive rate for the addition of a narrow component. We found that for the soft state, only one of these simulations was improved by more than ΔW = −17.5, and for the hard state, 23 out of the 5000 simulations were improved by more than ΔW = −10.6. In other words, we estimate a statistical significance of 3.7σ and 2.8σ for the addition of a narrow Fe component in the soft and hard state observations, respectively.

Finally, prompted by the clear presence of a hot thermal component, when modeling the soft state spectrum, we replaced relxilllp with the sum of nthcomp (Zdziarski et al. 1996; Życki et al. 1999), which models Comptonization of seed photons originating from a blackbody, and relxillNS (García et al. 2022), which models reflection of a blackbody spectrum rather than a power law. We tied the disk blackbody temperature to that of the seed photon temperature for both nthcomp and relxillNS, and we fixed the electron temperature of nthcomp, kT_e , at 1000 keV. We found that the fit was not affected by the type of seed blackbody specified for nthcomp, so for consistency, we used a disk blackbody seed spectrum. We found that this model improved the fit (ΔW = −10, Δd. o. f. = 1) when compared to the relxilllp model. We also tried modeling the hard state spectrum using a combination of a Comptonized disk blackbody with a reflected Fe line (diskbb + nthcomp + rellinelp). We found that this model struggled to describe the hard state spectrum (W/d. o. f. = 472/372), particularly at high photon energies (>20 keV). It resulted in a photon index of Γ = 1.8 and an electron temperature of kT_e = 9.1 keV.

Having arrived at a suitable model for the spectra during both the soft (diskbb + nthcomp + relxillNS + Gaussian) and hard (cutoffpl + relxilllp + Gaussian) states, we performed a Markov Chain Monte Carlo (MCMC) analysis in Xspec in order to explore the parameter space in detail. We used the Goodman-Weare algorithm (Goodman & Weare 2010) with a chain length of 10⁶ and a burn-in length of 10⁵. We initialized all walkers in a Gaussian distribution about the best-fit parameters, with σ defined by the covariance matrix resulting from least-squares fitting. All spectral parameters are therefore given as the median values of the final posterior distributions, and the errors represent the bounds between which 90% of the samples lie. ¹³ We found that both models recovered a high spin, a > 0.7, and an inner disk inclination angle around 25°. However there was significant degeneracy such that each model had valid solutions with different values for spin, disk inclination, inner disk radius, and other key parameters (see the Appendix for details regarding the individual fits). In order to break this degeneracy, we performed a joint fit wherein the inner disk inclination, the spin parameter, and the iron abundance were tied between the two spectral states.

Table 2 lists the median parameters for the joint fit resulting from the MCMC analysis described above, as well as the W-statistic of the best fit. Figure 5 shows the spectra as well as the best-fit models and residuals for both the soft and the hard states. By performing a joint fit, we successfully narrowed the parameter space, resulting in a consistent solution across both the soft and hard states with a spin of a = 0.967 ± 0.013 and an inclination of i = 264 ± 05.

Table 2. Spectral Parameters Determined by Performing a Joint Fit of the Soft and Hard State Spectra in Xspec

Model Component	Parameter	Soft State (NuObs 1)	Hard State (NuObs 2)
tbabs	NH (1022 cm−2)	${4.2}_{-0.1}^{+0.2}$	6.1 ± 0.2
Gauss	μ (keV)	6.8 ± 0.1	6.3 ± 0.1
	K (10−5Photon cm−2 s−1)	17.7 ± 0.4	${6.26}_{-0.07}^{+0.05}$
	Equivalent Width (eV)	42 ± 3	42 ± 1
diskbb	kTin (keV)	${1.38}_{-0.03}^{+0.02}$	⋯
	${({R}_{\mathrm{km}}/{D}_{10})}^{2}\cos i$	${6.4}_{-0.3}^{+0.2}$	⋯
nthcomp/cutoffpl	Γ	2.42 ± 0.04	${1.58}_{-0.02}^{+0.01}$
	Ecut (keV)	⋯	28.9 ± 0.4
	Norm (10−3)	9.6 ± 0.4	${21.3}_{-0.8}^{+0.7}$
relxillNS/relxilllp	h (Rg)	⋯	5.7 ± 0.2
	a	0.967 ± 0.013
	Inclination (deg)	26.4 ± 0.5
	Rin (Rg)	2.9 ± 0.2	${7.90}_{-0.16}^{+0.15}$
	$\mathrm{log}(\xi /\mathrm{erg}\,\mathrm{cm}\,{{\rm{s}}}^{-1})$	2.58 ± 0.06	${2.77}_{-0.02}^{+0.04}$
	AFe (Solar)	${0.77}_{-0.01}^{+0.02}$
	$\mathrm{log}(N)$ (cm−3)	<15.3	⋯
	frefl a	0.61 ± 0.01	${0.83}_{-0.03}^{+0.02}$
Fbol(10−10 erg cm−2 s−1) b		6.9 ± 0.1	2.85 ± 0.04
Lbol(1035 erg s−1) c		9.1 ± 0.1	3.73 ± 0.05
W/d. o. f.		834/695

Notes.

^a We determined the reflection fraction by first fitting with the direct emission components included in the relxill models (i.e., with a positive value of f_refl), then we determined the normalizations of the direct components by freezing the reflection fractions and separating the direct and reflected components (with a negative value of f_refl). ^b Unabsorbed bolometric (0.1–100 keV) flux. ^c Bolometric (0.1–100 keV) luminosity assuming a distance of 3.3 kpc.

Download table as:  ASCIITypeset image

The Fe Kα emission line profile (Figure 6) shows clear evolution of the region responsible for this emission, with the peak shifting toward lower energies by about 0.5 keV and the overall profile becoming more pronounced as the source evolved from the soft to the hard state. Interestingly, we found that the centroid of the narrow Fe line during the hard state was less than 6.4 keV at the 95% confidence level. In other words, we detect a redshift in the narrow component at a significance of about 2σ. We also fit the data while freezing the centroid at 6.4 keV given that this is still consistent with our measurement of μ = 6.3 ± 0.1 keV, but this resulted in a somewhat degraded fit (ΔW = 3; Δd. o. f. = 1), and it led to significantly looser constraints on parameters such as spin, inclination, and lamppost height.

The evolution of the accretion disk is further reflected in the best-fit relativistic reflection models, which suggest a significant increase in the inner radius of the accretion disk. This could constitute evidence for disk truncation in the hard state. In order to determine whether the choice of spectral model affects the apparent evolution of the inner disk radius, we performed a joint fit wherein the disk reflection features of both the soft and hard state spectra were modeled using relxilllp (rather than using relxillNS/relxilllp for the soft/hard states, respectively). For this fit, we tied the spin parameter, the inner disk inclination, and the Fe abundance between the soft and the hard states. This resulted in a slightly larger value for the inner disk radius in both states—R_in = 3.4 ± 0.2 R_g(where R_g = GM/c²) for the soft state and ${R}_{\mathrm{in}}={8.6}_{-1.1}^{+0.5}\,{R}_{{\rm{g}}}$ for the hard state—but the radii remained significantly different, indicating that the evolution of the inner disk radius that we observe is independent of the choice of relxillNS or relxilllp.

The narrow Fe line components may be interpreted as distant disk reflection. Therefore we investigated whether nonrelativistic reflection modeling could describe this component. We chose to use another member of the relxill suite, xillver (García & Kallman 2010; García et al. 2013), for this component. In this model, the relxill and xillver components can be interpreted as reflection from the inner and outer regions of the disk, respectively. In the soft state, distant reflection, using either xillver or xillverNS to model reflection of the power-law component or the blackbody component, respectively, is not able to fit the data as well as a simple Gaussian and results in a change in the fit statistic of ΔW = 10.

For the hard state, replacing the Gaussian emission line with the distant reflection model did not change the fit statistic appreciably, but we found that the resulting parameters show significant degeneracies, such as between the disk inclination, the power-law photon index, and the lamppost height. We tied the photon index, Γ, and the cutoff energy, E_cut, of the distantly reflected power law to those of the power law reflected from the inner disk, and we tied the iron abundance of the inner and outer disk, but we allowed the ionization of the two components to vary independently. Given that we measured a centroid of μ = 6.3 ± 0.1 keV, slightly lower than the rest-frame energy of the neutral Fe Kα line, when modeling the narrow emission component using a Gaussian component, we also allowed the redshift of the outer disk component represented by xillver to vary.

We found that this model was not sensitive to the inclination of the outer disk. When allowed to vary independently of the inner disk inclination, the model preferred a high outer disk inclination of ${i}_{\mathrm{out}}={79}_{-11}^{+6}$ deg (Δi ≡ i_out − i_in = 57° ± 10° ), but tying the inner and outer inclinations did not affect the fit statistic and resulted in an inclination that was closer to that resulting from the joint fits, $i={22}_{-5}^{+4}$ deg. In addition to a slightly smaller inclination angle, adding the xillver component while tying the inner and outer disk inclinations also resulted in an increased inner disk radius (${R}_{\mathrm{in}}={14}_{-5}^{+6}\,{R}_{{\rm{g}}}$).

We also found that allowing the redshift of the xillver component to vary improved the fit by ΔW = −4 compared to the same model with the redshift frozen at a value of zero. We measure a redshift of z < 3.4 × 10⁻² at the 99% confidence level, corresponding to an upper limit on the line-of-sight velocity of the emitting region of v < 10,000 km s⁻¹, assuming a rest-frame energy of 6.4 keV. We found that performing the same calculations using the posterior distribution of Gaussian centroids resulting from the joint fit produces upper and lower limits differing from these by less than 10%. The fit is slightly degraded by fixing the redshift to zero, but the main effect on the parameters is a decrease in the inner disk inclination by a few degrees. We note that the source is unlikely to experience a strong redshift due to the motion of the binary system, and it certainly does not experience cosmological redshift, but rather any observed strong redshift is more likely to be caused by motion of the emitting region itself. In Section 4.3, we discuss one scenario—that of ionized outflows—which could result in red- and blueshifting of narrow emission.

We investigated the source variability by producing cospectra (Bachetti et al. 2015) for each of the NuSTAR observations. We first corrected the photon times of arrival for the motion of NuSTAR using barycorr. We specified the Chandra source position reported by Chakrabarty et al. (2021), and we used the JPL planetary ephemeris DE-430. In order to minimize the impact of the motion of the source on the detectors on our timing analysis, we analyzed each of the CHU combinations separately. We split the event files into segments of length 1024 s, then binned the events with a time resolution of 2048⁻¹ s in order to probe variability between 10⁻³ Hz and 10³ Hz. ¹⁴ This range of frequencies allowed us to investigate both the low-frequency noise, and to search for any hint of pulsations potentially arising from an NS source. For each segment, we calculated the cospectrum between FPMA and FPMB. We found that for both observations, only CHU combination CHU12 contained continuous good time intervals of length greater than 1024 s, and therefore this was the only CHU combination for which we calculated cospectra. For each observation, we then averaged all of the cospectra. Finally, we applied deadtime and background corrections by multiplying the averaged fractional rms normalized power by ${\left(1-\tau (\overline{S+B})\right)}^{-2}{\left(\overline{S+B}/\overline{S}\right)}^{2}$, where τ = 25 ms is the deadtime per event, S and B are the source and background rates, respectively, calculated from the event list, and bars indicate the geometric mean of the FPMA and FPMB count rates.

Table 3 shows the timing properties of the source during both observations. Because the source landed on the gap between detectors during NuObs 1, which is sure to exacerbate any systematic variability due to motion of the source on the focal plane, we do not present the cospectrum here. The quality of the data was not sufficient to perform modeling of the cospectrum, but we did not observe a clear flattening in the low-frequency noise of the source. We were able to place an upper limit on the low-frequency noise of rms <8% by integrating the measured power in the range 10⁻³–10⁻¹ Hz. In order to estimate the continuum rms variability, we produced an averaged cospectrum with segment length 64 s (due to the shorter segment length, we were able to utilize the data from all CHU combinations except for CHU2) and sampled the resulting error distributions for the power in the range 0.1–10 Hz. We produced 10⁵ sample cospectra in this range, and calculated the total rms for each. The result was an upper limit, defined as the 99th percentile of this sample, of rms <15%.

The source did not fall on the chip gap during NuObs 2, and we therefore present the averaged, logarithmically rebinned cospectrum in Figure 7. Although there is clear red noise, similar to the previous observation, we do not observe a low-frequency turnover. We fitted the low-frequency (<0.1 Hz) noise to a power law and found an index of γ = −2.1 ±0.1. This is consistent with the tail of a zero-centered Lorentzian, which is often used to describe the low-frequency noise in accreting BHs and NSs (Belloni et al. 2002). We integrated the low-frequency noise and obtained rms = 12% ± 2%, and we obtained an upper limit on the power integrated between 0.1–10 Hz of rms <14% using the same method as described for the previous observation (in this case only CHU13 was excluded).

Zoom In Zoom Out Reset image size

Noting a potential feature in the cospectrum at ∼2 × 10⁻² Hz, we performed a search for quasiperiodic oscillations (QPO) by stepping logarithmically from 10⁻³ to 1024 Hz and fitting the unbinned hard state cospectrum to the power law described above plus a Lorentzian centered at each frequency step. For each frequency, we recorded the change in the χ² fit statistic compared to the power law alone. Using this method, we did not find any significant QPO candidates. To determine whether the potential feature could be due to coherent pulsations, we also folded the events for each observation, summing all CHU combinations, at 10,000 pulse frequencies between 2 × 10⁻² Hz and 3 × 10⁻² Hz and calculating the ${Z}_{1}^{2}$ statistic (Buccheri et al. 1983) for each candidate frequency. For the soft state observation, NuObs 1, we found a peak in the ${Z}_{1}^{2}$ distribution at ∼44 s with a significance of 2.1σ, and for the hard state observation, NuObs 2, the highest peak was at ∼42 s and had a significance of only 0.7σ. We further searched for pulsations in each observation by producing cospectra for the frequency range 0.008–2048 Hz, again using the sum of all CHU combinations. We did not observe any significant peaks during either observation.

While there is no "smoking gun" evidence for one particular type of compact object (for example, the detection of pulsations would prove the presence of an NS), the evidence we have presented thus far favors the case of a BH X-ray binary.

The soft state spectrum is described well by a Comptonized multitemperature disk blackbody with relativistic reflection features, which may represent returning radiation (see Connors et al. 2020, 2021; Lazar et al. 2021), and the hard state spectrum is described by a reflected power law alone. Joint modeling using relxilllp and relxillNS indicates that the accreting object has a high spin, a ≈ 0.97. Given the relation a = cJ/GM², where J is the angular momentum of the spinning compact object, and M is its mass, and assuming a moment of inertia $I=\tfrac{2}{5}{{MR}}^{2}$ for an NS, we may estimate the maximum spin that an NS can attain. Assuming a canonical NS with M = 1.4M_⊙ and R = 10 km, we arrive at a spin of $a\approx 0.4\tfrac{{\nu }_{\mathrm{spin}}}{\mathrm{kHz}}$, meaning that even the most rapidly rotating NSs will not achieve spin parameters approaching that which we have measured. We consider this a compelling piece of evidence that favors the classification of MAXI J1848015 as an accreting BH—explaining the broad Fe line in the case of an accreting NS would prove difficult.

Additionally, using archival Chandra observations of the GC01 cluster, Hare et al. (2021) were able to place an upper limit of ∼3.3 × 10³⁰ erg s⁻¹ on the quiescent luminosity in the 0.5–10 keV band. Whereas accreting NSs tend to have quiescent luminosities around 10³³ erg s⁻¹ (Tsygankov et al. 2017), the luminosity of MAXI J1848015 prior to outburst is more in line with those measured for BHs, perhaps due to the presence of an event horizon rather than a boundary layer (Garcia et al. 2001).

The initial outburst observed by MAXI was very short-lived compared to many BH X-ray binaries, which tend to show exponential decays with e-folding times of tens of days (McClintock & Remillard 2006); however, similarly rapid outbursts are not unheard of (see Maitra & Bailyn 2006). Of course, the count rate in the limited energy band does not always reflect the mass accretion rate, especially in cases where the source spectrum changes significantly. It is, however, interesting to note that some objects, e.g., gamma-ray bursts, exhibit power-law decays similar to the flux decrease observed for MAXI J1848015 by NuSTAR, MAXI, and Swift (Miller et al. 2021).

MAXI J1848015 reached an unabsorbed bolometric luminosity of ∼10³⁶ erg s⁻¹ during the soft state, and dropped to a luminosity of ∼4 × 10³⁵ erg s⁻¹ in the hard state. The general shape of the spectrum observed during each of these states is consistent with accretion onto an NS or a BH, and despite the low luminosities, observations of broad emission features around 6.4 keV indicate the presence of an optically thick accretion disk that extends close to the central accretor and reaches a high temperature of ∼1.4 keV. Models of accretion onto BHs predict the formation of an advection-dominated accretion flow, or ADAF, at mass accretion rates below a few percent of the Eddington limit, resulting in truncation of the disk at large radii (Esin et al. 1997). Indeed, observations have demonstrated that X-ray binaries tend to undergo state transitions at ∼2% of the Eddington limit (Maccarone 2003; Vahdat Motlagh et al. 2019). Given the inconsistency we observe between the low source luminosity and evidence for a small disk radius, we consider two effects that could suppress the observed luminosity: disk inclination and obscuration.

Reflection features—the broad Fe line profile and excess "Compton hump" above 10 keV—allowed us to constrain the inclination of the inner regions of the disk as well as the inner disk radius. Reflection modeling of the soft state spectrum resulted in an inner disk radius of R_in = 2.9 ± 0.2R_g. Even for an NS accretor with mass M = 1.4M_⊙, this implies an inner disk radius of about 6 km, whereas the flux of the disk blackbody implies an effective inner disk radius of ${R}_{\mathrm{in}}\sqrt{\cos i}=0.8\,\mathrm{km}$, assuming a distance to the source of 3.3 kpc. Several effects, including gravitational redshift, Compton scattering of disk photons (Kubota & Makishima 2004), spectral hardening (Shimura & Takahara 1995), and boundary condition corrections (Kubota et al. 1998), will increase or decrease the actual value of the inner disk radius compared to the apparent radius, but in total, these effects are unlikely to introduce a factor of more than order unity.

Thus we are left with a contradiction between inner disk radius measurements, which could presumably be resolved by assuming a high inclination, i, of the inner disk, resulting in a smaller projected disk flux from the point of view of an observer. While reflection modeling appears to rule this out, preferring an inclination between 20° and 30°, it has been shown that the inclination inferred from reflection modeling can conflict drastically from the actual inclination of a system, possibly due to obscuration of the blue wing of the broad Fe line (Connors et al. 2019).

High inclination of the inner disk alone may not be able to explain the low flux of MAXI J1848015. We therefore also consider obscuration of the inner regions of the disk by the outer disk. In this scenario, the outer regions of the disk intercept the line of sight between the observer and the inner disk. If the outer regions of the disk have a clumpy structure, this will result in an effective partial coverer, and may be consistent with the change in column density we have measured between the soft and hard states. Alternatively, we may be observing emission from the inner disk, which has been scattered off the outer regions of the disk on the side opposite the observer.

A combination of these two effects has been proposed for the high-mass X-ray binary, V4641 Sgr (Koljonen & Tomsick 2020), which has similarly been shown to exhibit evidence for a disk extending close to the central BH despite a low Eddington ratio (Miller et al. 2002). In fact, the hard state of MAXI J1848015 shares a number of similar characteristics with V4641 Sgr. Both exhibit power-law spectra with indices between 1 and 2 and relatively low cutoff energies (<30 keV). Additionally, both systems undergo short outbursts with durations of tens of days or less (Wijnands & van der Klis 2000; Revnivtsev et al. 2002), and on shorter timescales, they both exhibit relatively featureless power spectra with red noise components described by a power law of slope ∼ −2 (Maitra & Bailyn 2006). However, the amplitude of variability in the 0.1–10 Hz range that we have observed, while poorly constrained, is somewhat lower than that of V4641 Sgr (Maitra & Bailyn 2006). In both of these systems, reprocessing of emission from the central regions of the accretor could help to explain this behavior by "smearing out" variability on short timescales.

Given the low luminosity as well as the very short outburst duration, we also suggest the possibility that the source may reside in a compact binary. A smaller-than-usual disk could lead to short outbursts, and depending upon the orbital period, the mass accretion rate may be low (Deloye & Bildsten 2003). It is interesting to note that BH binaries undergoing mass transfer in globular clusters are likely to harbor both low-mass donors and relatively low-mass BHs (Kremer et al. 2018). Because we are unable to detect the orbital period, and due to the fact that a compact binary would imply a small companion, which is contradicted by the potential NIR counterpart of the system, we consider this case speculative. However, given the rarity of known BHs in compact binaries, it is an interesting possibility nonetheless and one that warrants further observations of the source.

We found that the narrow Fe lines observed in the soft and hard state spectra have different centroid energies, evolving from ∼6.8 keV in the soft state to ∼6.3 keV in the hard state. This change likely corresponds to a change in the ionization of the emitting region, perhaps originating from the outer regions of the accretion disk, with neutral or lowly ionized Fe i–xvii responsible for the line in the hard state and highly ionized Fe xxv–xxvi responsible for the line in the soft state (García et al. 2013). Indeed, the significantly higher thermal X-ray flux in the Fe Kα complex energy range during the soft state would lead to a higher ionization.

Rather than an evolution of the emitting region, it is also possible that narrow emission from neutral Fe was actually present during both states, but that it did not increase drastically in flux during the soft state and was therefore not detectable above the thermal-dominated continuum, which was significantly higher in flux around 6.4 keV as compared with the hard state. Indeed, when MAXI J1848015 was observed by NICER during a second, less pronounced increase in flux, Miller et al. (2021) not only reported that the spectrum was described well by a blackbody with temperature kT = 1.0 keV, indicating that the source had again entered the soft state, but the authors also reported both the detection of a prominent, narrow emission line at 6.7 keV with flux K = 1.5 × 10⁻⁵ photon cm⁻² s⁻¹, as well as a tentative detection of a weaker line at 6.4 keV.

In order to test whether the lower-energy line could have been present but undetected during the NuSTAR observation of the soft state, we added a second Gaussian component to the soft state in our joint spectral model. We froze the centroid at μ = 6.4 keV and the width at σ = 10⁻⁵ keV. We allowed the flux of this component to vary, then we calculated the 99% confidence upper limit as defined by a change in fit statistic of ΔW = 6.63. The resulting upper limit was K < 7.7 × 10⁻⁵ photon cm⁻² s⁻¹. In other words, we cannot rule out the possibility that neutral Fe emission was present in the soft state at a similar flux level to that observed in the hard state.

One alternative scenario to that of differing ionization is an ionized outflow. Given that we do not observe clear absorption features that might indicate disk winds, such an outflow may instead take the form of a jet launched during the short outburst observed by MAXI. In this case, the narrow emission lines may originate from fast-moving, ionized blobs of plasma on opposite sides of the accretor along the jet axis. One would then expect to see a pair of components, one of which represents the blueshifted jet component moving toward the observer, and the second of which originates from the redshifted jet component moving away from the observer. Due to relativistic beaming, the blueshifted component would have a higher observed flux than that of the redshifted component. Additionally, depending on the orientation of the system, one may expect a delay between the appearance of the blue component and that of the red component due to the difference in light-travel time from each of the emitting blobs. Migliari et al. (2002) observed a pair of Fe lines during Chandra observations of SS 433, which they were able to spatially associate with extended jet emission. They determined that the Fe line emitting regions were located at a distance of >10¹⁷ cm from the central accretor. For a source with low or moderate inclination, this would result in a light-travel time between the two lobes of 30–40 days.

Indeed, we have shown that the 6.8 keV line had a higher flux than, and was observed prior to, the 6.3 keV line. If the two lines originated from regions with similar ionization states, and the difference in energy is solely due to blue- and redshifting, then we may calculate the velocity of the narrow line emitting regions:

$$\begin{eqnarray}&&\displaystyle \frac{{E}_{\mathrm{obs}}}{{E}_{\mathrm{rest}}}=\displaystyle \frac{1}{\gamma \left(1-\beta \cos \theta \right)}\end{eqnarray} \tag{ 1 }$$

where E_obs is the energy of the line in the frame of the observer, E_rest is the energy of the line in the rest frame of the emitting region, β is the ratio of the velocity of the emitting region to the speed of light, γ is the Lorentz factor, and θ is the angle at which the emitter is moving with respect to the line of sight. If we assume that the velocities of the red and blue components are equal in magnitude, the result is $\beta \cos \theta \approx 0.038$. Given that we have measured a moderate inclination via spectral analysis, we arrive at a deprojected velocity of β ≈ 0.043 and a rest-frame energy of E_rest ≈ 6.55 keV, which corresponds to Fe XXII–XXIII.

Assuming a delay of about a week in the appearance of the redshifted line, we arrive at a distance of ∼10¹⁶ cm between the central accretor and the narrow line emitting regions. This distance is difficult to square with a velocity of 0.04c in the case that the putative jets were launched during the initial outburst of MAXI J1848015, but as we have shown, the 6.3 keV line may have gone undetected early in the outburst. While the evidence we have presented may be consistent with emission from an outflow, we are unable to meaningfully distinguish this case with the simpler case of different ionization states. The jet model may be probed by future observations of the source by observatories, such as Chandra, with high spatial resolution and high spectral resolution in the Fe Kα complex.