betaFeedback

WorksPhysics and Astronomy

The Disk Awakens: IXPE's Sequel Observation of GS 1354-64 Suggests Changing-Look Behavior in the Steep-Power-Law State

Abstract

After decades of only failed transitions, the black hole X-ray binary GS 1354-64 has undergone a complete outburst again. We present its first X-ray polarimetric investigation in the fading steep-power-law (SPL) state, following an earlier intermediate-state IXPE campaign during the same outburst. The 2-8 keV emission has a polarization degree (PD) of 2.3 +/- 0.2% and polarization angle (PA) of 5 +/- 3 degrees. Compared with the intermediate state, where PD=3.9 +0.3/-0.2% and PA=-2 +/-2 degrees, the broadband PD is lower while the PA remains consistent. The intermediate-state type-C quasi-periodic oscillation state has also disappeared, marking a major reconfiguration of the accretion flow. Assuming a distance of 7 kpc, the observed the intensity drop from 2.5% to 1.6% of the Eddington luminosity over 10.2 days is punctuated by several broad hour-to-day-scale troughs, tracing two distinct tracks in hardness-intensity space. Simultaneous spectroscopy shows the troughs' spectra contains a cooler disk with smaller apparent normalization, while no change in the Compton component's photon index or normalization is required between tracks. The model-independent analysis identifies a steeper increase in PD with energy outside the troughs, yielding G_E=1.5 +/- 0.2%/keV, compared with G_E=0.7 +0.3/-0.5%/keV during the troughs. Their duration and spectropolarimetric morphology distinguish the troughs from traditional short absorption dips seen in X-ray binaries, providing a possible stellar-mass analog of the recurrent accretion changes that drive "changing-look" behavior in supermassive black hole systems.

Email: [email protected]

keywords: Polarimetry (1278); Black holes (162); X-ray binary stars (1811); Accretion (14); X-ray Astronomy (1810)

Introduction

Black hole X-ray binaries (BHXBs) serve as dynamic laboratories of accretion, changing their luminosity, spectral shape, and rapid variability as they traverse a sequence of characteristic X-ray states [71, 18]. During a canonical outburst, a source generally rises into a luminous hard state dominated by Comptonized emission from a hot inner flow or corona, passes through one or more intermediate states, and enters into a thermal-dominant soft state or a steep power-law (SPL; historically, very-high) state [53, 71, 18]. The SPL state is characterized by the coexistence of substantial thermal-disk emission and a steep Comptonized tail (commonly with photon index Γ≳2.4\Gamma\gtrsim2.4) together with substantially weaker broadband variability than in the hard and hard-intermediate states. The SPL state can exhibit QPOs and moderate rapid variability, but its timing properties span a broader range than those of the thermal-dominant state [53, 71].

The hard-to-soft transition may be accompanied by transient relativistic ejections, and once the source is in a disk-bright state, broadband X-ray variability is generally reduced and the compact, steady radio jet is quenched. Changes in the geometry of the accretion disk and corona are also inferred [25, 29, 71, 18]. Although this broad phenomenology is well established, how the inner disk and corona reconfigure during state transitions and what drives substantial variability within nominally stable states remain uncertain.

The Imaging X-ray Polarimetry Explorer (IXPE) [79, 93] provides a new view of this evolution by measuring the linear polarization of X-rays in the 2–8 keV band. The polarization degree (PD) and polarization angle (PA) depend on the geometry and orientation of the emitting and scattering regions, providing complementary diagnostics to X-ray spectroscopy and timing. In hard and intermediate states, the IXPE band is often dominated by Comptonized emission whose polarization can constrain the shape and orientation of the corona [39, 89, 31, 24].

In SPL states, IXPE probes radiation emerging from the optically thick accretion disk alongside a substantial steep Comptonized component, making the observed polarization signal sensitive both to the intrinsic disk signal and its dilution or modification by coronal scattering. Classical models of thermal disk polarization assume a plane-parallel, pure electron-scattering atmosphere with photon sources at its base [15, 78, 70]. In this case, the PD depends primarily on inclination. However, absorption in partially ionized plasma and distributed photon sources can significantly alter this picture [42, 46, 87]. General and special relativistic effects modify the local polarization signatures by rotation of PA, resulting in PD reduction [81, 16, 19]. Because higher-energy disk photons preferentially originate at smaller radii, these effects are expected to produce energy-dependent polarization signatures that encode the structure of the innermost disk. Black hole spin influences this signal by setting the location of the innermost stable circular orbit (ISCO) and therefore dictating the magnitude of the PD reduction and PA rotation effects, as well as controlling the strength of returning radiation—photons that are bent back onto the disk and scatter before reaching the observer [76, 77]. X-ray polarimetry in disk-bright soft and SPL states can therefore complement continuum-fitting and reflection measurements of black-hole spin while testing assumptions about disk thickness, atmospheric scattering, coronal geometry, and alignment of the innermost disk [52, 72, 54].

The growing sample of disk-bright BHXBs observed by IXPE reveals remarkable diversity. In 4U 1957+115, a soft-state polarization signal at the few-percent level has been interpreted in the context of relativistic disk models as evidence for a rapidly spinning black hole under the assumed system geometry [49]. The soft-state transient 4U 1630−47 exhibited a much larger PD, approaching ∼8%\sim8\%, together with a pronounced increase with energy, motivating interpretations involving a highly inclined disk, an extended scattering atmosphere, potential departures from a standard geometrically thin disk, and considerable disk-wind contributions to the observed polarimetric properties [66]. A subsequent IXPE observation of 4U 1630−47 in the SPL state likewise measured a large PD that increased strongly with energy while the PA remained approximately constant, demonstrating that pronounced energy-dependent polarization can persist when substantial thermal-disk emission coexists with a steep Comptonized component [75]. The low-inclination LMC X-1 produced an upper limit of 1% PD when observed by IXPE, while the higher-inclination LMC X-3 showed a low, few-percent PD consistent with an energy-independent PA, and consistent with relativistic disk calculations for its low spin [85]. These observations demonstrate how the range of polarimetric morphology present among disk-bright sources can reflect differences in inclination, disk structure, returning radiation, and geometry of the residual Comptonized component.

Multi-epoch IXPE observations have also begun to trace how polarization evolves across accretion-state transitions. In the persistent BHXB Cygnus X-1, the hard state shows a PD of approximately 4% with a PA aligned with the radio jet, favoring a corona extended in the plane of the disk [39]. In the soft state, the PD decreases while the PA remains broadly stable, consistent with an increasing contribution from thermal disk emission [83, 38]. Similarly, during the 2023–2024 outburst of Swift J1727.8−1613, IXPE followed the source from the hard and intermediate states into the soft state and subsequently toward the returning hard state [89, 86, 31, 63]. Its PD fell from ∼ 3–4% in the hard state to below ∼ 1% as the thermal disk came to dominate, before the hard-state PD returned during the soft-to-hard transition. IXPE additionally observed GX 339−4 during the soft-intermediate and soft states of its 2023–2024 outburst, detecting polarization during the soft-intermediate observation but obtaining only an upper limit in the soft state [50]. These campaigns establish that PD responds sensitively to state evolution, but often lack insight into the PA evolution across state transition, as significant disk-bright polarization has not previously been reported for a transient BHXB that was also observed polarimetrically in a harder state of the same outburst.

The Galactic transient GS 1354−64 (BW Cir) offers an unusual setting in which to examine disk-bright accretion with X-ray polarimetry. Optical dynamical measurements yield a mass function of f(M)≃5.7 M⊙f(M) \simeq5.7\,M_\odot, identifying the compact companion as a black hole [12]. Although several studies favor a comparatively high orbital inclination, potentially near 75°, both the viewing geometry and distance remain poorly constrained [13, 23, 26, 44, 20]. In particular, recent work favors a distance of 7.1−3.9+4.87.1^{+4.8}_{-3.9} kpc [62], although published estimates have ranged from 0.7−0.3+3.40.7^{+3.4}_{-0.3} kpc to ∼ 20–25 kpc [13, 26].

Much of the previous observational history of GS 1354−64 was defined by its reluctance to enter softer, disk-bright states. After the luminous 1987 discovery outburst [36], its 1997 and 2015 outbursts remained dominated by hard or hard-intermediate emission and were classified as “failed” or “stalled” transitions [84, 37, 60] (C. Brock­sopp et al. 2001). These episodes showed strong Comptonization and, at some epochs, relativistic reflection, but yielded model-dependent and disparate conclusions concerning the location of the inner disk and the black-hole spin [84, 37] (C. Brock­sopp et al. 2001). The eventual emergence of substantial thermal-disk emission during the 2025–2026 outburst therefore opened an accretion regime that had rarely been accessible in this source. The first IXPE observation during the 2025–2026 outburst (taken 2026 February 07–09) sampled GS 1354−64 between episodes of bright X-ray flaring, while the source occupied an intermediate state [56, 57, 69]. The observation yielded a significant polarization detection, with a 2–8 keV PD of 3.9−0.2+0.3%3.9^{+0.3}_{-0.2}\% at PA = −2∘±2∘-2^\circ\pm2^\circ, and a PD that increased strongly with energy at gE=1.3±0.2%g_E = 1.3 \pm0.2\% keV−1^{-1}, while the PA remained approximately constant [69]. It also revealed a low-frequency quasi-periodic oscillation (QPO) near 4–5.5 Hz. These properties were consistent with an accretion flow in which a structured Comptonizing region continued to contribute substantially to the 2–8 keV emission [69]. The source subsequently evolved into a substantially less variable, disk-bright configuration while retaining a steep Comptonized tail. This, together with the disappearance of the type-C QPO detected during the preceding intermediate-state observation [69], motivates its classification as a fading SPL state and allows IXPE to revisit the same transient after a major reconfiguration of its accretion flow.

Here, we present the first IXPE observation of GS 1354−64 in a fading SPL state and compare its polarization with that measured previously during the intermediate state. In Section 2, we describe the IXPE and contemporaneous NuSTAR observations, data reduction, and supporting X-ray monitoring. In Section 3, we establish the source’s position within the 2025–2026 outburst, characterize its model-independent polarization properties, and perform spectral and spectropolarimetric analyses of the two distinct intensity tracks. In Section 4, we combine these results to investigate whether the broad troughs arise from changes in the thermal disk, the relative disk–corona contribution, or intervening material. We compare their spectral and polarimetric morphology with dipping X-ray binaries and consider whether the recurrent transitions between luminosity configurations provide a phenomenological stellar-mass analog of changing-state behavior across the black-hole mass scale.

Observations and Data Reduction

Observations

MAXI Monitoring

We monitored the evolution of GS 1354−64 throughout its 2025–2026 outburst using per-orbit MAXI light curves in the 2–4, 4–10, and 10–20 keV bands. The selected interval extends from 2025 December 29 to 2026 April 10 (MJD 61038–61141), encompassing the source’s transition into the SPL state.

IXPE

IXPE observed GS 1354−64 in the SPL state during 2026 March 28–April 09 (ObsID 05250399), providing an effective exposure of approximately 450 ks for each Detector Unit (DU). We follow the reduction procedure adopted for the earlier IXPE observation of the source ([69]). Source events are extracted from the Level 2 files using a circular region with a radius of 60″, positioned with the “centroid” tool in SAOImage DS9 ([33]). Given the high source count rate of 2.32 s−1 arcmin−22.32\ \mathrm{s}^{-1}\,\mathrm{arcmin}^{-2} per DU, we apply neither background rejection nor background subtraction, following [17]. We adopt the time-dependent response version 20260101 and group the spectra into 200 eV bins, corresponding to an oversampling of the instrumental energy resolution by a factor of approximately 2–3.

NuSTAR

GS 1354−64 was observed by the Nuclear Spectroscopic Telescope Array (NuSTAR) on 2026 March 29 (ObsID 81202310002), contemporaneously with the IXPE observation, as indicated by the gray shaded interval in Figure 1. The observation provides a total effective exposure of ∼21\sim21 ks for each focal plane module (FPM). We produce cleaned event files using nupipeline with CALDB version 20260903 and adopt the standard screening criteria, including saomode=optimized and tentacle=yes. Source events are extracted from circular regions with radii of 120″, centered in SAOImage DS9 as described above ([33]). For each detector, we select a 150″-radius background region from a source-free portion of the field. Spectra and response files are generated using nuproducts. We optimally group the FPMA and FPMB spectra following [34], additionally requiring at least 30 counts per bin, and analyze them over the 3–79 keV energy range. Because this observation was obtained after the development of the NuSTAR FPMB multilayer-insulation (MLI) tear, we included the energy-dependent FPMB correction using the multiplicative table model nuMLIv1.mod.2 We first determined the FPMB-to-FPMA cross-normalization using energies above 7 keV, where the MLI correction is negligible, and then fixed this constant while fitting the FPMB MLI covering fraction over the full 3–79 keV band. The corresponding FPMA MLI covering fraction was fixed to unity.

Figure 1. MAXI/GSC light curve of GS 1354−64 during its 2025–2026 outburst (MJD 61038–61141). The dashed blue box in the 2–20 keV light curve depicts the intermediate-state IXPE observation (05002101) reported in S. Ravi et al. [69], the solid purple box depicts the SPL IXPE observation (05250399), and the gray shading denotes the contemporaneous NuSTAR observation (81202310002). The middle and top panels show the SPL IXPE 2–8 keV light curve and 4–8/2–4 keV hardness ratio, respectively, calculated from the combined event lists of all three DUs and binned at 50 s and 3 ks, respectively. Pink shading marks the time intervals identified with the lower-track troughs.

Data Reduction

Model-Independent Polarimetric Analyses

We characterize the model-independent polarization using two complementary approaches. First, we apply the weighted PCUBE algorithm implemented in ixpeobssim version 33.0.0, adopting the SIMPLE weighting prescription3 ([35, 8]). We additionally perform event-level Bayesian inference with QUEEN-BEE version 3.3, using the constant-polarization model scout_const ([67, 68]).

Results

IXPE Observation

As shown in Figure 1, the second IXPE observation sampled the late decline of the 2025–2026 outburst, after the source had faded substantially from its earlier X-ray flares. During the observation, the IXPE 2–8 keV count rate declined from approximately 20 to 10 counts s−1\mathrm{s}^{-1} while exhibiting pronounced, structured variability. Several broad troughs interrupt the otherwise comparatively smooth decline, appearing to transition between quasi-stable flux levels. These troughs are accompanied by increases in the 4–8 keV/2–4 keV hardness ratio. We therefore divide the IXPE observation into two tracks: the upper track consists of “declining-continuum” intervals, and the lower track of “trough” intervals. The lower-track intervals are highlighted in pink in Figure 1. The MAXI/GSC HID in Figure 2 is suggestive of a separation between the two track selections, with the IXPE HID in Figure 3 resolving them into distinct regions of hardness–intensity space separate from both each other and the previous intermediate-state IXPE observation.

Figure 2. MAXI/GSC HID of GS 1354−64 from MJD 55053 to 61141, using 12-hour bins. Green points illustrate the ongoing 2025–2026 outburst. Blue starred points correspond to the intermediate-state IXPE observation (05002101) reported in S. Ravi et al. [69], while purple and pink starred points correspond to the upper and lower tracks of the SPL IXPE observation (05250399), respectively. Data from MJD 58490–58570 are excluded to avoid contamination from the nearby transient MAXI J1348−630.

Figure 3. IXPE HID of GS 1354−64 during the intermediate- and SPL observations, using 3.5 ks bins. Hardness is defined as the ratio of the summed DU1–3 count rates in the 4–8 and 2–4 keV bands, while intensity is the summed DU1–3 count rate in the 2–8 keV band. Blue starred points correspond to the intermediate-state IXPE observation (05002101) reported in S. Ravi et al. [69], while purple and pink points correspond to the upper and lower tracks of the SPL IXPE observation (05250399), respectively.

We examined each nominal trough boundary using the summed 2–8 keV light curves from the three IXPE DUs. The light curves were constructed independently with 50- and 100-s bins, and Poisson uncertainties were calculated from the number of events in each bin. Within a ±0.03-day interval around each boundary, we fitted a logistic transition in which the low- and high-state count rates, transition midpoint, and characteristic timescale τ\tau were allowed to vary. We defined the transition duration as the time required for the fitted count rate to change between 10% and 90% of its full amplitude, t10−90=2ln⁡(9)τt_{10-90} = 2 \ln(9)\tau, with uncertainties propagated from the fit covariance. Most boundaries coincide with gaps in the IXPE coverage and are therefore unresolved or only weakly constrained. The best-sampled boundary is the egress of the second broad trough, for which the nearest bracketing bins are separated by only 50 and 100 s in the two light curves. The 50- and 100-s fits give t10−90=0.3±0.1t_{10-90} = 0.3 \pm0.1 and 0.6±0.20.6 \pm0.2 ks, respectively, with transition midpoints agreeing to within approximately 25 s. The precise duration is therefore sensitive to the adopted binning, but both fits indicate that the egress occurred within approximately 5–11 minutes, more than two orders of magnitude faster than the ∼105\sim10^{5}-s trough duration.

The presence of substantial thermal-disk emission together with a steep Comptonized tail in the spectrum, combined with the marked reduction in broadband rms variability relative to the preceding intermediate-state IXPE observation (2026 February 07–09), supports the interpretation that GS 1354−64 had entered a fading SPL state (see Sect. 3.3,3.4); the concurrent disappearance of the type-C QPO provides additional evidence for this transition. Formally, the Comptonized component contributes ∼66% of the unabsorbed 2–20 keV continuum, exceeding the 50% threshold used to identify the SPL state when no QPO is detected, per R. A. Remillard & J. E. McClintock [71]. While the time-averaged

photon index, Γ≃2.3\Gamma\simeq2.3, lies slightly below the conventional Γ>2.4\Gamma> 2.4 criterion for SPL, the spectrum steepens to Γ=2.8±0.3\Gamma= 2.8 \pm0.3 during the final 100 ks of the observation, formally crossing this threshold and motivating our interpretation of the source as evolving into a fading SPL state in this work.

Polarimetry

We detect significant 2–8 keV polarization from the source during the SPL IXPE observation using both model-independent polarimetric approaches. PCUBE measures a PD of 2.3±0.2%2.3 \pm0.2\% and a PA of 5∘±3∘5^\circ\pm3^\circ (1σ1\sigma uncertainties), corresponding to a significance of 9.7σ9.7\sigma. The QUEEN-BEE scout_const analysis (Figure 4) returns polarization parameters consistent with the PCUBE results, with a PD of 2.3−0.3+0.2%2.3^{+0.2}_{-0.3}\% and a PA of 4∘±3∘4^\circ\pm3^\circ (90% credible intervals). Comparison of the constant-polarization model with the null unpolarized hypothesis yields a log Bayes factor of Δln⁡(Z)=108±1\Delta\ln(Z) = 108 \pm1.

Figure 4. Polarimetric summary of both GS 1354−64 IXPE observations. Left: Constant-PA (scout_const) QUEEN-BEE polar plot. The intermediate-state observation (05002101) is colored blue and the full SPL observation (05250399) is colored purple. Contours represent 68%, 95%, and 99.7% credible intervals. Right: Energy-dependent PD and PA behavior of intermediate-state IXPE data (05002101) in blue and full SPL IXPE data (05250399) in purple. Binned PCUBE results are shown as individual points. Error bars denote 1σ1\sigma uncertainties. Points in grey denote bins with <3σ<3\sigma significance. Unbinned QUEEN-BEE results are shown as solid lines. Shading denotes asymmetric 90% credible intervals. The corresponding absorbed disk fraction—defined as the ratio of disk flux to total (disk + Comptonized) flux—in each 1-keV bin across the 2–8 keV band, is overplotted in yellow.

The energy-binned PCUBE analysis (right panel of Figure 4) reveals a significant increase in PD with energy, while the PA remains constant across the IXPE band. Consistent with this result, the QUEEN-BEE scout_energy_evol model finds decisive evidence, per the modified Jeffreys scale (H. Jeffreys [32]), for PD energy dependence at a rate of GE=1.2−0.2+0.1 %/keV\mathcal{G}_{E} = 1.2^{+0.1}_{-0.2}\ \%/\mathrm{keV}. Comparing this model, which assumes a linear increase in PD with energy across the IXPE bandpass, with the constant-polarization model yields Δln⁡(Z)=28±1\Delta\ln(Z) = 28 \pm1.

In contrast, the QUEEN-BEE scout_energy_rot analysis finds no evidence for energy-dependent PA rotation. The comparison with the constant-PA model yields Δln⁡(Z)=−4±1\Delta\ln(Z) = -4 \pm1. The inferred rotation rate is ωE=−0.3−2.1+2.7 ∘ keV−1\omega_{E} = -0.3^{+2.7}_{-2.1}\,^\circ\,\mathrm{keV}^{-1}, consistent with zero (Ravi et al., in prep.). We likewise find no significant temporal rotation of the PA using the QUEEN-BEE scout_rot analysis. The time-rotating and constant-PA models yield Δln⁡(Z)=−3±1\Delta\ln(Z) = -3 \pm1.

We also performed track-resolved QUEEN-BEE analyses of the upper and lower tracks (Figure 6). For the upper track, the scout_const analysis gives PD=2.0±0.3%\mathrm{PD} = 2.0 \pm0.3\% and PA=3−4+5∘\mathrm{PA} = 3^{+5}_{-4}{}^\circ, while the lower-track analysis gives PD=2.8±0.4%\mathrm{PD} = 2.8 \pm0.4\% and PA=5−4+5∘\mathrm{PA} = 5^{+5}_{-4}{}^\circ. The constant-polarization model is decisively favored over the unpolarized hypothesis in both intervals, with Δln⁡(Z)=40.9±0.8\Delta\ln(Z) = 40.9 \pm0.8 and 59.7±0.559.7 \pm0.5 for the upper and lower tracks, respectively.

Figure 6. Constant-PA (scout_const) QUEEN-BEE polar plot. The intermediate-state observation (05002101) is colored blue. The upper-track SPL observation (05250399) is colored purple while the lower track SPL observation is colored pink. Contours represent 68%, 95%, and 99.7% credible intervals.

The track-resolved scout_energy_evol analyses give GE,upper=1.5±0.2% keV−1\mathcal{G}_{E,\mathrm{upper}} = 1.5 \pm0.2\%\,\mathrm{keV}^{-1} and GE,lower=0.7−0.5+0.3% keV−1\mathcal{G}_{E,\mathrm{lower}} = 0.7^{+0.3}_{-0.5}\%\,\mathrm{keV}^{-1}, with log Bayes factors of Δln⁡Z=26±1\Delta\ln Z = 26 \pm1 and 1.2±0.91.2 \pm0.9, respectively, relative to the corresponding constant-polarization models. The scout_energy_rot analyses similarly give ωE,upper=3±4∘ keV−1\omega_{E,\mathrm{upper}} = 3 \pm4^\circ\,\mathrm{keV}^{-1} and ωE,lower=−2−3+4∘ keV−1\omega_{E,\mathrm{lower}} = -2^{+4}_{-3}{}^\circ\,\mathrm{keV}^{-1}, with Δln⁡Z=−4.3±0.7\Delta\ln Z = -4.3 \pm0.7 and −4.5±0.6-4.5 \pm0.6, respectively, relative to the constant-PA models. Finally, the scout_time_rot analyses give ωt,upper=−2±1∘ day−1\omega_{t,\mathrm{upper}} = -2 \pm1^\circ\,\mathrm{day}^{-1} and ωt,lower=2±3∘ day−1\omega_{t,\mathrm{lower}} = 2 \pm3^\circ\,\mathrm{day}^{-1}, with Δln⁡Z=−1.6±0.6\Delta\ln Z = -1.6 \pm0.6 and −4.0±0.6-4.0 \pm0.6, respectively. Thus, energy-dependent PD is strongly favored only for the upper track (Figure 5), while neither track shows evidence for energy- or time-dependent PA rotation.

Figure 5. Energy-dependent PD for the upper and lower tracks of the SPL IXPE observation. The purple squares and pink circles show the energy-binned PCUBE measurements for the upper and lower tracks, respectively; measurements below 3σ3\sigma significance are shown in gray. The solid curves show the corresponding QUEEN-BEE scout_energy_evol results, with shaded regions denoting the asymmetric 90% credible intervals.

The PD of each SPL track differs from that measured during the intermediate-state observation (S. Ravi et al. [69]) at greater than 99% posterior credibility (Figure 6). In contrast, the upper- and lower-track polarization posteriors overlap substantially: although the lower track has a marginally higher best-fitting PD, the two tracks are not statistically distinct from one another. Their PAs are likewise consistent with both each other and the intermediate-state observation.

Spectroscopy

NuSTAR spectral fitting

We first characterized the broadband continuum using the simultaneous NuSTAR observation 81202310002. The FPMA and FPMB spectra were fit in the 3–

79 keV band with the model constant*nuMLI*TBabs(diskbb+nthComp*smedge+laor) using XSPEC version 12.14.0h [1]. Interstellar absorption was modeled with TBabs, adopting the abundances of J. Wilms et al. [94] and the photoelectric cross sections of D. A. Verner et al. [90]. The hydrogen column density was frozen at NH=7×1021 cm−2N_{\mathrm{H}} = 7 \times10^{21}\ \mathrm{cm}^{-2}, consistent with the value adopted in the analysis of the preceding intermediate-state observation S. Ravi et al. [69]. The NuMLI covering fraction was fixed to unity for FPMA and allowed to vary for FPMB. The resulting 3–79 keV spectra had net count rates of 35.90 and 33.08 counts s−1^{-1} for FPMA and FPMB, respectively, yielding ∼1.43×106\sim1.43 \times10^{6} net source counts across the two FPMs.

The diskbb component K. Mitsuda et al. [55] describes the multitemperature thermal emission from the accretion disk, while nthComp P. T. Życki et al. [96] represents thermal Comptonization of the disk photons. The seed-photon temperature of nthComp was tied to the diskbb temperature, a disk seed spectrum was assumed by fixing inp_type = 1, and the redshift was fixed at zero. The electron temperature was unconstrained and reached the upper boundary of the allowed range, kTe=1000kT_{\mathrm{e}} = 1000 keV, which we retained in the subsequent IXPE analysis. This boundary value should not be interpreted as a physical measurement of the coronal temperature: at relativistic electron temperatures, the approximations underlying nthComp no longer provide an accurate thermal-Comptonization spectrum and can produce substantial discrepancies from more exact calculations A. Niedźwiecki et al. [58]. Moreover, an observed power-law-like tail in this regime may more plausibly arise from Comptonization by a nonthermal or hybrid electron distribution J. Poutanen & I. Vurm [65]. We therefore use nthComp only as a phenomenological description of the continuum within the observed bandpass and do not assign physical significance to the fitted electron temperature or other inferred coronal plasma properties.

The FPMA multiplicative constant was fixed to unity. We determined the FPMB-to-FPMA cross-normalization from a fit excluding the FPMB data below 7 keV, obtaining CB=1.00163C_{\mathrm{B}} = 1.00163, and subsequently froze this value while fitting the full band with the MLI correction. The resulting FPMB MLI covering fraction was fMLI,B=0.89±0.02f_{\mathrm{MLI,B}} = 0.89 \pm0.02. The best-fitting parameters are reported in Table 1. The corrected NuSTAR continuum was characterized by a disk temperature of kTin=0.804−0.010+0.009kT_{\mathrm{in}} = 0.804^{+0.009}_{-0.010} keV and a photon index of Γ=2.32±0.01\Gamma= 2.32 \pm0.01. All reported spectral parameter uncertainties correspond to 90% confidence.

ComponentParameterNuSTARIXPE averageUpper trackLower track
CONSTANTFPMA or DU11f1^{\mathrm{f}}1f1^{\mathrm{f}}1f1^{\mathrm{f}}1f1^{\mathrm{f}}
FPMB/FPMA1.00163f1.00163^{\mathrm{f}}⋯\cdots⋯\cdots⋯\cdots
DU2/DU1⋯\cdots1.002−0.007+0.0061.002^{+0.006}_{-0.007}1.004±0.0021.004 \pm0.0020.999±0.0030.999 \pm0.003
DU3/DU1⋯\cdots0.979−0.006+0.0070.979^{+0.007}_{-0.006}0.977±0.0020.977 \pm0.0020.982±0.0030.982 \pm0.003
NUMLIFPMA covering fraction1f1^{\mathrm{f}}⋯\cdots⋯\cdots⋯\cdots
FPMB covering fraction0.89±0.020.89 \pm0.02⋯\cdots⋯\cdots⋯\cdots
TBABSNH (1022 cm−2)N_{\mathrm{H}}\ (10^{22}\ \mathrm{cm}^{-2})0.7f0.7^{\mathrm{f}}0.7f0.7^{\mathrm{f}}0.7f0.7^{\mathrm{f}}0.7f0.7^{\mathrm{f}}
DISKBBkTin (keV)kT_{\mathrm{in}}\ (\mathrm{keV})0.804−0.010+0.0090.804^{+0.009}_{-0.010}0.68±0.010.68 \pm0.010.70±0.010.70 \pm0.010.63±0.010.63 \pm0.01
Normalization350±20350 \pm20470±10470 \pm10252±5252 \pm5228±7228 \pm7
NTHCOMPΓ\Gamma2.32±0.012.32 \pm0.012.3±0.12.3 \pm0.12.4±0.22.4 \pm0.22.4±0.12.4 \pm0.1
kTe (keV)kT_{\mathrm{e}}\ (\mathrm{keV})1000f1000^{\mathrm{f}}1000f1000^{\mathrm{f}}1000f1000^{\mathrm{f}}1000f1000^{\mathrm{f}}
kTbb (keV)kT_{\mathrm{bb}}\ (\mathrm{keV})=kTin= kT_{\mathrm{in}}=kTin= kT_{\mathrm{in}}=kTin= kT_{\mathrm{in}}=kTin= kT_{\mathrm{in}}
Seed type1f1^{\mathrm{f}}1f1^{\mathrm{f}}1f1^{\mathrm{f}}1f1^{\mathrm{f}}
Normalization0.267−0.009+0.0100.267^{+0.010}_{-0.009}0.22±0.030.22 \pm0.030.12±0.020.12 \pm0.020.12−0.01+0.020.12^{+0.02}_{-0.01}
SMEDGEEedge (keV)E_{\mathrm{edge}}\ (\mathrm{keV})7.35−0.09+0.107.35^{+0.10}_{-0.09}7.35f7.35^{\mathrm{f}}7.35f7.35^{\mathrm{f}}7.35f7.35^{\mathrm{f}}
τmax\tau_{\mathrm{max}}1.26−0.07+0.081.26^{+0.08}_{-0.07}1.26f1.26^{\mathrm{f}}1.26f1.26^{\mathrm{f}}1.26f1.26^{\mathrm{f}}
Index−2.67f-2.67^{\mathrm{f}}−2.67f-2.67^{\mathrm{f}}−2.67f-2.67^{\mathrm{f}}−2.67f-2.67^{\mathrm{f}}
Width (keV)7f7^{\mathrm{f}}7f7^{\mathrm{f}}7f7^{\mathrm{f}}7f7^{\mathrm{f}}
LAOREline (keV)E_{\mathrm{line}}\ (\mathrm{keV})6.7f6.7^{\mathrm{f}}6.7f6.7^{\mathrm{f}}6.7f6.7^{\mathrm{f}}6.7f6.7^{\mathrm{f}}
Emissivity index3f3^{\mathrm{f}}3f3^{\mathrm{f}}3f3^{\mathrm{f}}3f3^{\mathrm{f}}
Rin (GM/c2)R_{\mathrm{in}}\ (GM/c^2)1.235f1.235^{\mathrm{f}}1.235f1.235^{\mathrm{f}}1.235f1.235^{\mathrm{f}}1.235f1.235^{\mathrm{f}}
Rout (GM/c2)R_{\mathrm{out}}\ (GM/c^2)400f400^{\mathrm{f}}400f400^{\mathrm{f}}400f400^{\mathrm{f}}400f400^{\mathrm{f}}
Inclination (deg)81−2+181^{+1}_{-2}81f81^{\mathrm{f}}81f81^{\mathrm{f}}81f81^{\mathrm{f}}
Normalization (10−310^{-3})1.0±0.21.0 \pm0.21.0f1.0^{\mathrm{f}}1.0f1.0^{\mathrm{f}}1.0f1.0^{\mathrm{f}}
Gain, DU1Slope⋯\cdots0.989130f0.989130^{\mathrm{f}}0.989130f0.989130^{\mathrm{f}}0.989130f0.989130^{\mathrm{f}}
Offset (keV)⋯\cdots0.0445967f0.0445967^{\mathrm{f}}0.0445967f0.0445967^{\mathrm{f}}0.0445967f0.0445967^{\mathrm{f}}
Gain, DU2Slope⋯\cdots0.960±0.0020.960 \pm0.0020.959655f0.959655^{\mathrm{f}}0.959655f0.959655^{\mathrm{f}}
Offset (keV)⋯\cdots0.109±0.0090.109 \pm0.0090.109062f0.109062^{\mathrm{f}}0.109062f0.109062^{\mathrm{f}}
Gain, DU3Slope⋯\cdots0.996±0.0030.996 \pm0.0030.996152f0.996152^{\mathrm{f}}0.996152f0.996152^{\mathrm{f}}
Offset (keV)⋯\cdots0.04±0.010.04 \pm0.010.042862f0.042862^{\mathrm{f}}0.042862f0.042862^{\mathrm{f}}
DISKBBF2−8 (erg cm−2s−1)F_{2-8}\ (\mathrm{erg\ cm}^{-2}\mathrm{s}^{-1})1.006×10−91.006 \times10^{-9}5.060×10−105.060 \times10^{-10}3.379×10−103.379 \times10^{-10}1.558×10−101.558 \times10^{-10}
NTHCOMPF2−8 (erg cm−2s−1)F_{2-8}\ (\mathrm{erg\ cm}^{-2}\mathrm{s}^{-1})1.040×10−91.040 \times10^{-9}7.109×10−107.109 \times10^{-10}3.767×10−103.767 \times10^{-10}3.401×10−103.401 \times10^{-10}
TotalF2−8 (erg cm−2s−1)F_{2-8}\ (\mathrm{erg\ cm}^{-2}\mathrm{s}^{-1})2.051×10−92.051 \times10^{-9}1.222×10−91.222 \times10^{-9}7.201×10−107.201 \times10^{-10}5.014×10−105.014 \times10^{-10}
Disk fraction Fdisk/(Fdisk+FnthComp)F_{\mathrm{disk}}/(F_{\mathrm{disk}} + F_{\mathrm{nthComp}})0.4920.4920.4160.4160.4730.4730.3140.314
χ2/ν\chi^2/\nu470.34/385470.34/385107.98/77107.98/7781.55/8181.55/81107.59/81107.59/81

Table 1. Spectral parameters obtained from the NuSTAR, time-averaged IXPE, and independently fitted track-resolved spectra. Errors are reported at the 90% confidence level.

Following the treatment adopted for the preceding intermediate-state IXPE-NuSTAR joint observation S. Ravi et al. [69], we included LAOR A. Laor [40] to describe the broad Fe-K emission-line feature and SMEDGE K. Ebisawa [22] to describe the associated absorption-edge-like curvature. For LAOR, we adopted the default maximal-spin Kerr geometry a∗=0.998a_* = 0.998, corresponding to the Thorne limit, fixed the line energy at 6.7 keV for Fe XXV, and assumed an emissivity index of q=3q = 3. The inner and outer radii were fixed at 1.235 GM/c21.235\,GM/c^2 (as expected in the maximal spin case, motivated by H. Liu et al. [45] and 400 GM/c2400\,GM/c^2, respectively).4 For SMEDGE, the edge width and index were fixed at 7 keV and −2.67-2.67, respectively. The remaining Fe-K parameters were allowed to vary, and their best-fitting values are listed in Table 1. These components provide a consistent phenomenological description of the Fe-K structure but should not be interpreted as a self-consistent reflection model.

The best-fitting NuSTAR spectrum and residuals are shown in Figure 7. Using statistical uncertainties alone, the MLI corrected fit gave χ2/ν=470.34/385\chi^2/\nu= 470.34/385, indicating that some residual structure remains.5 We nevertheless use this fit to establish a consistent phenomenological treatment of the disk and Compton continuum and Fe-K emission for the subsequent IXPE analysis, rather than as a fully optimized physical description of the spectrum. Adding an energy-independent systematic uncertainty of 1.05% per spectral bin reduces the fit statistic to χ2/ν=383.21/385\chi^2/\nu= 383.21/385, indicating that the adopted phenomenological model reproduces the NuSTAR spectral shape to approximately the percent level.

Figure 7. Simultaneous fit to the NuSTAR FPMA and FPMB spectra over 3–79 keV. The upper panel shows the unfolded spectra, total model, and the DISKBB and NTHCOMP*SMEDGE continuum contributions. The weak LAOR component is included in the total model but falls below the displayed range. The lower panel shows the residuals in units of the measurement uncertainty. The phenomenological model describes the overall broadband continuum and Fe-K structure but leaves some statistically significant residual structure.

Time-averaged IXPE intensity spectrum

We next fit the time-averaged Stokes-II spectra from the three IXPE detector units over the 2–8 keV band using the same spectral model. Starting from the NuSTAR solution, we thawed the diskbb temperature and normalization, the nthComp photon index, and the nthComp normalization. These parameters were refit because the source continuum need not remain identical between the NuSTAR and IXPE spectra, particularly across the two identified light-curve tracks highlighted in Figure 1, and because the time-averaged IXPE intensity spectrum provides the initial continuum decomposition and detector calibration for the track-resolved analysis. The hydrogen column density and electron temperature remained fixed at NH=7×1021 cm−2N_{\mathrm{H}} = 7 \times10^{21}\ \mathrm{cm}^{-2} and kTe=1000kT_{\mathrm{e}} = 1000 keV. The SMEDGE and LAOR parameters were frozen at the values obtained from the simultaneous NuSTAR spectral fitting.

The spectral-shape parameters were tied across the three detector units, and multiplicative constants accounted for their relative normalizations. The DU1 constant was fixed to unity, while the fitted DU2 and DU3 constants were 1.002−0.007+0.0061.002^{+0.006}_{-0.007} and 0.979−0.006+0.0070.979^{+0.007}_{-0.006}, respectively.

We first performed a time-averaged calibration fit in which independent linear gain corrections were allowed for the three detector units. The resulting gain slopes were 0.9891, 0.9596, and 0.9963 for DU1, DU2, and DU3, respectively, with corresponding offsets of 0.0446, 0.1086, and 0.0406 keV. To avoid degeneracy between the absolute gain scale and the continuum parameters, DU1 was subsequently adopted as the reference detector and fixed at its calibration-fit values, while only the DU2 and DU3 gain corrections were allowed to refine. The final gain slopes were 0.960 ±\pm 0.002 for DU2 and 0.996 ±\pm 0.003 for DU3, with corresponding offsets of 0.109 ±\pm 0.009 and 0.04 ±\pm 0.01 keV, respectively. These values were frozen in the track-resolved spectral and spectropolarimetric fits. The gain corrections account for residual relative energy-scale differences among the IXPE detector units and do not materially affect the inferred continuum.

The best-fitting time-averaged IXPE spectral parameters are listed in Table 1. The spectrum is characterized by a disk temperature of kTin=0.68±0.01kT_{\mathrm{in}} = 0.68 \pm0.01 keV and a photon index of Γ=2.3±0.1\Gamma= 2.3 \pm0.1. The fit gave χ2/ν=107.98/77\chi^{2}/\nu= 107.98/77. Because the time-averaged spectrum combines intervals with substantial intensity and spectral variability, this likely contributes to the remaining residuals and fit statistic. We nevertheless retain this phenomenological model because its primary purpose is to establish the intensity decomposition and detector calibration used in the subsequent spectropolarimetric analysis.

To quantify the secular decline across the observation, we additionally fit the first and final 100 ks intervals using the same model, with the Fe-K and detector-gain parameters fixed as described above. During the first 100 ks, the spectrum had kTin=0.73±0.02kT_{\mathrm{in}} = 0.73 \pm0.02 keV and Γ=2.3±0.3\Gamma= 2.3 \pm0.3, with χ2/ν=61.69/81\chi^{2}/\nu= 61.69/81. During the final 100 ks, the disk cooled to kTin=0.65±0.02kT_{\mathrm{in}} = 0.65 \pm0.02 keV while the Comptonized continuum steepened to Γ=2.8±0.3\Gamma= 2.8 \pm0.3, with χ2/ν=91.08/81\chi^{2}/\nu= 91.08/81. Extrapolating the fitted models over 0.01–100 keV and removing foreground absorption gives model-dependent bolometric fluxes of 5.37×10−95.37 \times10^{-9} and 3.42×10−93.42 \times10^{-9} erg cm−2^{-2} s−1^{-1}, respectively, corresponding to a decline of approximately 36%.

Track-resolved spectroscopy

We divided the observation into the same upper and lower tracks identified in Figure 1 and initially fit their Stokes-II spectra separately. For both tracks, the DISKBB temperature and normalization and the NTHCOMP photon index and normalization were free to vary. The absorption column, electron temperature, Fe-K parameters, and detector gain corrections were fixed as described above. The similarity between the trough separation and the binary orbital period motivates consideration of absorption as a possible contributor. We subsequently test whether allowing the neutral absorption column to differ between the upper and lower tracks, either alone or together with changes in the continuum normalizations, can explain their spectral differences.

The separate fits by track, reported in Table 1, show that the upper-track spectrum has a higher disk temperature than the lower track, 0.70±0.010.70 \pm0.01 versus 0.63±0.010.63 \pm0.01 keV. The upper-track fit gave χ2/ν=81.55/81\chi^{2}/\nu= 81.55/81, while the lower-track fit gave 107.59/81107.59/81, with the largest contribution in the latter arising from DU3 (χ2=50.07\chi^{2} = 50.07 for 29 bins). Relative to the lower track, the upper track has an approximately 12% higher disk temperature and 11% higher apparent disk normalization, whereas the independently fitted photon indices differ by less than 0.03 and the NTHCOMP normalizations by less than 1%. We tested this interpretation more systematically through joint fitting.

To determine which of these parameters were statistically required to differ, we next simultaneously fitted the six Stokes-II spectra from the upper and lower tracks, comprising three DU spectra per track, using a sequence of nested linking schemes. In every test, the three detector spectra within a given track shared the same spectral parameters. The DU2 and DU3 cross-normalizations were allowed to vary but were tied between the two tracks.

The detector gain corrections were frozen at the calibration values determined from the initial time-averaged IXPE fit. For the preferred fit described below (Case 3 in Table 2), in which the disk temperature and normalization differ between tracks while the Comptonization parameters remain tied, the resulting DU2 and DU3 cross-normalizations were 1.002±0.0021.002 \pm0.002 and 0.979±0.0020.979 \pm0.002, respectively. The absorption column, electron temperature, seed type, redshift, SMEDGE parameters, and LAOR parameters remained fixed at the values described above. Before examining changes in the intrinsic continuum, we tested an absorption-only Case 0 in which all continuum parameters were tied between the two tracks. Allowing a single common column density to vary gave χ2/ν=189729.7/167\chi^{2}/\nu= 189729.7/167. Allowing NHN_{\mathrm{H}} to differ between the tracks reduced the statistic to 28403.9/16628403.9/166, with nominal best-fitting values of NH,upper=2.4×1022N_{\mathrm{H,upper}} = 2.4 \times10^{22} cm−2^{-2} and NH,lower=4.6×1022N_{\mathrm{H,lower}} = 4.6 \times10^{22} cm−2^{-2}. Despite this improvement, the fit remained grossly unacceptable and drove the tied continuum to extreme compensating values. Allowing the absorption column and both component normalizations to differ between tracks, while retaining a common disk temperature and photon index, gave χ2/ν=250.20/164\chi^{2}/\nu= 250.20/164. This description remained statistically unacceptable and performed worse than the subsequently described thermal-evolution model (209.78/166209.78/166), despite having two additional free parameters. Variable neutral absorption therefore, even when combined with component-normalization changes, cannot adequately reproduce the track-resolved spectra without spectral-shape evolution.

ParameterCase 0Case 1Case 2Case 3Case 4Case 5
Parameters allowed to differNHN_{\mathrm{H}}NonekTinkT_{\mathrm{in}}kTin,NdiskkT_{\mathrm{in}}, N_{\mathrm{disk}}kTin,Ndisk,NnthkT_{\mathrm{in}}, N_{\mathrm{disk}}, N_{\mathrm{nth}}kTin,Ndisk,Nnth,ΓkT_{\mathrm{in}}, N_{\mathrm{disk}}, N_{\mathrm{nth}}, \Gamma
NHN_{\mathrm{H}} (1022 cm−210^{22}\ \mathrm{cm}^{-2})2.4−0.2+0.32.4^{+0.3}_{-0.2} 4.6−0.2+0.34.6^{+0.3}_{-0.2}0.70 (fixed)0.70 (fixed)0.70 (fixed)0.70 (fixed)0.70 (fixed)
kTinkT_{\mathrm{in}} (keV)0.36±0.020.36 \pm0.020.653±0.0090.653 \pm0.0090.618±0.0060.618 \pm0.0060.632±0.0090.632 \pm0.009 0.703±0.0070.703 \pm0.0070.63±0.010.63 \pm0.01 0.703±0.0070.703 \pm0.0070.70±0.010.70 \pm0.01
NdiskN_{\mathrm{disk}}9−3+6×1039^{+6}_{-3} \times10^3236±4236 \pm4254±4254 \pm4224±5224 \pm5 255±4255 \pm4228±7228 \pm7 253±5253 \pm5228±7228 \pm7
Γ\Gamma3.22−0.05+0.063.22^{+0.06}_{-0.05}2.48±0.092.48 \pm0.092.33±0.062.33 \pm0.062.36±0.082.36 \pm0.082.39±0.092.39 \pm0.09 2.4±0.22.4 \pm0.22.4±0.12.4 \pm0.1
NnthCompN_{\mathrm{nthComp}}0.74−0.09+0.100.74^{+0.10}_{-0.09}0.13±0.010.13 \pm0.010.112±0.0080.112 \pm0.0080.11±0.010.11 \pm0.010.11±0.010.11 \pm0.01 0.12±0.010.12 \pm0.010.12±0.020.12 \pm0.02
χ2/ν\chi^2/\nu28403.89/16628403.89/166189730.8/168189730.8/168341.60/167341.60/167209.78/166209.78/166207.68/165207.68/165207.66/164207.66/164
Null probability004.46×10−144.46 \times10^{-14}0.01210.01360.0119

Table 2. Comparison of simultaneous fits to the upper-track and lower-track Stokes-II spectra

We subsequently considered five progressively less restrictive linking schemes, summarized in Table 2. Case 1 represents the null hypothesis that the intrinsic continuum is unchanged between the tracks. Case 2 allows only the disk temperature to differ, and Case 3 additionally allows the DISKBB normalization to differ. Case 4 then allows the NTHCOMP normalization to vary. Finally, Case 5 allows both the NTHCOMP normalization and photon index to differ and is therefore the

Note— The upper- and lower-track columns show the results obtained when the two tracks were fit independently. Superscripts ff and pp denote parameters that were fixed or pegged at a fitting boundary, respectively. The IXPE gain corrections were fit using the time-averaged spectrum and then frozen in the independently fitted track-resolved spectra. The independently fitted values are reported as a diagnostic. The spectral decomposition subsequently frozen in the spectropolarimetric analysis was taken from Case 3 of Table 2. The reported component and total fluxes are absorbed 2–8 keV model fluxes. The disk fraction is defined over 2–8 keV from the absorbed fluxes as Fdisk/(Fdisk+FnthComp)F_{\mathrm{disk}}/(F_{\mathrm{disk}} + F_{\mathrm{nthComp}}) and excludes the weak LAOR contribution. All three IXPE fits contain 87 spectral bins; the track-resolved fits have four additional degrees of freedom because the DU2 and DU3 gain slopes and offsets were fixed to the values obtained from the time-averaged spectrum.

simultaneous-fit counterpart of the independently fitted spectra described above, with the detector calibration parameters treated consistently between tracks. This nested sequence determines which of the differences permitted in the independent fits are actually required by the data.

The results of the five simultaneous fitting cases are summarized in Table 2. In Case 1, all continuum parameters were tied between the upper and lower tracks.

This highly restrictive model gave χ2/ν=189730.8/168\chi^2/\nu= 189730.8/168 and was decisively rejected. Because this model forces the two tracks to have identical component normalizations as well as identical spectral shapes, its rejection primarily reflects the large intensity difference used to define the two tracks. Allowing only the disk temperature to vary in Case 2 provided a substantially improved, though still statistically unacceptable, fit with χ2/ν=341.60/167\chi^2/\nu= 341.60/167. Allowing the apparent disk nor­malization malization to vary in Case 3 further improved the fit by Δχ2=131.82\Delta\chi^2 = 131.82 for one additional free parameter.

Case 3 gave χ2/ν=209.78/166\chi^2/\nu= 209.78/166. In this model, the upper track had kTin=0.703±0.007kT_{\mathrm{in}} = 0.703 \pm0.007 keV and Ndisk=255±4N_{\mathrm{disk}} = 255 \pm4, compared with kTin=0.632±0.009kT_{\mathrm{in}} = 0.632 \pm0.009 keV and Ndisk=224±5N_{\mathrm{disk}} = 224 \pm5 in the lower track. The two tracks shared Γ=2.36±0.08\Gamma= 2.36 \pm0.08 and NnthComp=0.11±0.01N_{\mathrm{nthComp}} = 0.11 \pm0.01.

Allowing the NTHCOMP normalization to differ in Case 4 improved the fit relative to Case 3 by only Δχ2=2.10\Delta\chi^2 = 2.10 for one additional parameter. Additionally freeing the photon index in Case 5 changed the statistic by only Δχ2=0.02\Delta\chi^2 = 0.02 for one additional parameter. Thus, even the joint-fit counterpart of the independently fitted spectra improves upon Case 3 by only Δχ2=2.12\Delta\chi^2 = 2.12 for two additional parameters. The apparent agreement of the Comptonized continua in the independent fits is therefore confirmed quantitatively by the nested simultaneous comparison.

We adopt Case 3 as the preferred description of the track-resolved spectra. It is the simplest model that captures the principal spectral differences, and the addition of track-dependent Comptonization parameters does not produce a statistically meaningful improvement. In the adopted model, the upper track has an approximately 11% higher disk temperature and an approximately 14% higher apparent disk normalization than the lower track. The spectral evolution is therefore primarily associated with the thermal component, while the data require no independent change in the NTHCOMP photon index or normalization.7 The absorbed 2–8 keV flux decreased from 7.208×10−107.208 \times10^{-10} to 5.011×10−105.011 \times10^{-10} erg cm−2^{-2} s−1^{-1} between the upper and lower tracks. The corresponding disk flux decreased from 3.431×10−103.431 \times10^{-10} to 1.612×10−101.612 \times10^{-10} erg cm−2^{-2} s−1^{-1}, while the NTHCOMP flux decreased more modestly from 3.722×10−103.722 \times10^{-10} to 3.344×10−103.344 \times10^{-10} erg cm−2^{-2} s−1^{-1}. The disk therefore contributes 48.0% and 32.5% of the modeled continuum flux in the upper and lower tracks, respectively. Approximately 83% of the total track-to-track flux difference is associated with the thermal component. The preferred fit and its improvement relative to Case 2 are illustrated in Figure 8.

Figure 8. Simultaneous fits to the track-resolved IXPE Stokes-II spectra. The upper and lower tracks are shown using solid lines with filled points and dashed lines with open points, respectively. The upper panel shows the unfolded spectra and preferred Case 3 model, in which the DISKBB temperature and normalization were allowed to differ between the tracks, while the NTHCOMP photon index and normalization were tied. The thermal disk and Comptonized continuum are shown separately, while the weak LAOR component is included in the total model but falls below the displayed range. The middle panel shows the residuals for Case 2, in which only the disk temperature was allowed to differ, while the lower panel shows the residuals for Case 3. Allowing the apparent disk normalization to differ removes the systematic residual structure and improves the fit by Δχ2=131.82\Delta\chi^2 = 131.82 for one additional free parameter.

Spectropolarimetry

Following the Stokes-II analysis, we froze the spectral decomposition, cross-normalization constants, and detector gain parameters and fit the corresponding IXPE Stokes-QQ and UU spectra. The Stokes spectra were fit over 2–8 keV using the appropriate 20260101 modulation response files. Each track was fit separately, with the polarization parameters tied among detector units.

We tested seven polarization prescriptions, summarized in Table 3. These comprise a single energy-dependent POLLIN or energy-independent POLCONST component applied to the entire continuum; separate polarization components for the disk and Comptonized emission, with their polarization angles constrained to be orthogonal; an unpolarized disk combined with either constant or linearly energy-dependent Compton polarization; and a model in which the disk polarization degree was fixed at 6% and oriented orthogonally to the Note— The six cases are ordered from the absorption-only test through the most flexible model. In Case 0, the intrinsic continuum was tied while the absorption column was allowed to differ between the tracks. For parameters allowed to differ, values are listed as upper track/lower track except in the χ2/ν\chi^2/\nu row. A single value indicates that the parameter was tied between the tracks. For Cases 1–5, the absorption column was fixed. The electron temperature, seed type, redshift, smedge parameters, and laor parameters were fixed. The detector cross-normalizations were allowed to vary but were tied between the tracks, while the detector gain corrections were frozen at their time-averaged values. Case 3 was adopted for the subsequent spectropolarimetric analysis. Uncertainties are reported at 90% confidence; those for the strongly disfavored restrictive cases are formal confidence intervals conditional on the corresponding model.

CasePolarization prescriptionPDdisk\mathrm{PD}_{\mathrm{disk}} (%)PDC\mathrm{PD}_{\mathrm{C}} or A1,CA_{1,\mathrm{C}} (%)AslopeA_{\mathrm{slope}} or Aslope,CA_{\mathrm{slope,C}} (% keV−1^{-1})PA (deg)χ2/ν\chi^2/\nu
Upper Track
1Common POLCONSTPD=1.7±0.4\mathrm{PD}=1.7\pm0.4⋯\cdots1±81\pm8192.84/172
2Common POLLINA1<0.6A_1<0.60.8±0.20.8\pm0.20±60\pm6178.69/171
3Orthogonal disk + POLLIN corona7±67\pm619±1419\pm14−3±3-3\pm31±61\pm6174.53/170
4Orthogonal disk + POLCONST corona2±22\pm26±26\pm2⋯\cdots0±60\pm6176.72/171
5Unpolarized disk + POLCONST corona0 (fixed)4.2±1.04.2\pm1.0⋯\cdots1±71\pm7181.33/172
6Unpolarized disk + POLLIN corona0 (fixed)A1,C=2−2+3A_{1,\mathrm{C}}=2^{+3}_{-2}0.8±0.80.8\pm0.80±60\pm6178.63/171
76% disk + orthogonal POLCONST corona6 (fixed)10.6±1.010.6\pm1.0⋯\cdots−1±6-1\pm6191.85/172
Lower Track
1Common POLCONSTPD=2.4±0.5\mathrm{PD}=2.4\pm0.5⋯\cdots1±71\pm7158.16/172
2Common POLLINA1=2±1A_1=2\pm10.1±0.60.1\pm0.61±71\pm7158.02/171
3Orthogonal disk + POLLIN corona<7<75−2+105^{+10}_{-2}−0.6−2.0+0.9-0.6^{+0.9}_{-2.0}1±71\pm7158.06/170
4Orthogonal disk + POLCONST corona<1.4<1.43.8−0.9+1.03.8^{+1.0}_{-0.9}⋯\cdots−1−6+7-1^{+7}_{-6}159.36/171
5Unpolarized disk + POLCONST corona0 (fixed)3.8±0.93.8\pm0.9⋯\cdots−1±7-1\pm7159.36/172
6Unpolarized disk + POLLIN corona0 (fixed)A1,C=5±2A_{1,\mathrm{C}}=5\pm2−0.6±0.9-0.6\pm0.91±71\pm7158.06/171
76% disk + orthogonal POLCONST corona6 (fixed)7.2±0.97.2\pm0.9⋯\cdots−6±7-6\pm7180.30/172

Table 3. Track-resolved spectropolarimetric fits. Polarization degrees and POLLIN amplitudes are given in percent, and polarization slopes are given in percent per keV. The reported PA is either the common continuum PA or the PA of the Comptonized component, as appropriate. Uncertainties and upper limits are reported at the 90% confidence level.

Comptonized component.8 The weak laor component was treated as unpolarized in all cases.

For pollin, the polarization degree is parameterized as

PD(E)=A1+(E−1 keV)Aslope,(1)\mathrm{PD}(E) = A_1 + (E - 1\ \mathrm{keV})A_{\mathrm{slope}}, \tag*{(1)}

where A1A_1 is the extrapolated polarization degree at 1 keV and AslopeA_{\mathrm{slope}} is its change per keV. We fixed the PA slope to zero because neither the model-independent analysis nor the spectropolarimetric fits indicated significant PA evolution. Because 1 keV lies below the fitted bandpass, A1A_1 and AslopeA_{\mathrm{slope}} are strongly correlated. In the principal simultaneous track comparison, we therefore allowed A1A_1 to take negative values, treating it only as an algebraic intercept, while verifying that the modeled PD remained positive throughout the observed 2–8 keV band.

For the upper track, the common pollin model gave χ2/ν=178.69/171\chi^2/\nu= 178.69/171, improving upon the common polconst fit by Δχ2=14.15\Delta\chi^2 = 14.15 for one additional parameter. The extrapolated 1-keV polarization reached its lower boundary, with A1<0.6%A_1 < 0.6\% at 90% confidence, while Aslope=0.8±0.2% keV−1A_{\mathrm{slope}} = 0.8 \pm0.2\%\,\mathrm{keV}^{-1}. The corresponding PA was

0±6∘0 \pm6^\circ. The upper-track data therefore favor a polarization degree that increases with energy.

The component-resolved models provided similarly acceptable descriptions. The lowest fit statistic was obtained for orthogonally polarized disk and Comptonized components with an energy-dependent Compton polarization, yielding χ2/ν=174.53/170\chi^2/\nu= 174.53/170. However, the disk and Compton polarization parameters were strongly covariant. Models containing an unpolarized disk and a polarized Comptonized component were also acceptable. With energy-independent Compton polarization, the inferred coronal polarization was PDC=4.2±1.0%\mathrm{PD}_{\mathrm{C}} = 4.2 \pm1.0\% at PAC=1±7∘\mathrm{PA}_{\mathrm{C}} = 1 \pm7^\circ and gave χ2/ν=181.33/172\chi^2/\nu= 181.33/172. Allowing the coronal PD to vary linearly with energy improved the fit to χ2/ν=178.63/171\chi^2/\nu= 178.63/171, with A1,C=2−2+3%A_{1,\mathrm{C}} = 2^{+3}_{-2}\%, Aslope,C=0.8±0.8% keV−1A_{\mathrm{slope,C}} = 0.8 \pm0.8\%\,\mathrm{keV}^{-1}, and PAC=0±6∘\mathrm{PA}_{\mathrm{C}} = 0 \pm6^\circ.

For the lower track, the common pollin and polconst models gave nearly identical fit statistics, χ2/ν=158.02/171\chi^2/\nu= 158.02/171 and 158.16/172158.16/172, respectively. The common polconst model yielded PD=2.4±0.5%\mathrm{PD} = 2.4 \pm0.5\% at PA=1±7∘\mathrm{PA} = 1 \pm7^\circ, demonstrating that an energy-independent polarization adequately describes this track. The common pollin fit gave A1=2±1%A_1 = 2 \pm1\%, Aslope=0.1±0.6% keV−1A_{\mathrm{slope}} = 0.1 \pm0.6\%\,\mathrm{keV}^{-1}, and PA=1±7∘\mathrm{PA} = 1 \pm7^\circ.

The lower-track component-resolved fits were again degenerate. In the orthogonal constant-polarization model, the disk polarization reached zero, with a 90% upper limit of PDdisk<1.4%\mathrm{PD}_{\mathrm{disk}} < 1.4\%. With the disk fixed as unpolarized, the Comptonized component had PDC=3.8±0.9%\mathrm{PD}_{\mathrm{C}} = 3.8 \pm0.9\% at PAC=−1±7∘\mathrm{PA}_{\mathrm{C}} = -1 \pm7^\circ, giving χ2/ν=159.36/172\chi^2/\nu= 159.36/172. Allowing its polarization to vary linearly with energy gave χ2/ν=158.06/171\chi^2/\nu= 158.06/171, with A1,C=5±2%A_{1,\mathrm{C}} = 5 \pm2\%, Aslope,C=−0.6±0.9% keV−1A_{\mathrm{slope,C}} = -0.6 \pm0.9\%\,\mathrm{keV}^{-1}, and PAC=1±7∘\mathrm{PA}_{\mathrm{C}} = 1 \pm7^\circ.

Overall, the upper-track data favor an increase in net PD with energy, whereas the lower track is adequately described by an energy-independent polarization. The component-resolved fits do not uniquely distinguish an intrinsic polarization gradient from energy-dependent dilution by the disk and Comptonized components.

To test the apparent difference in the energy dependence of the PD directly, we also fit the upper- and lower-track Stokes-QQ and UU spectra simultaneously with POLLIN. Because A1A_1 is an extrapolated intercept below the fitted bandpass, we allowed it to take negative values while requiring the resulting PD to remain positive throughout the 2–8 keV band. In the null model, the two tracks were allowed to have different polarization intercepts but were constrained to share a common PD slope and PA. This model gave χ2/dof=340.01/344\chi^2/\mathrm{dof} = 340.01/344. Allowing the PD slopes to differ improved the fit to χ2/dof=334.82/343\chi^2/\mathrm{dof} = 334.82/343, corresponding to Δχ2=5.19\Delta\chi^2 = 5.19 for one additional free parameter.

The separate-slope fit gave A1=−0.9±1.2%A_1 = -0.9 \pm1.2\% and Aslope=1.2±0.5% keV−1A_{\mathrm{slope}} = 1.2 \pm0.5\%\,\mathrm{keV}^{-1} for the upper track, and A1=2.3±1.5%A_1 = 2.3 \pm1.5\% and Aslope=0.1±0.6% keV−1A_{\mathrm{slope}} = 0.1 \pm0.6\%\,\mathrm{keV}^{-1} for the lower track. The common PA was 0∘±5∘0^\circ\pm5^\circ. Although the upper-track intercept is negative, the modeled PD remains positive throughout the observed band, reaching approximately 0.35% at 2 keV; the negative value therefore reflects only the extrapolation of the linear model to 1 keV. Under the usual likelihood-ratio interpretation, the improvement gives a nominal p=0.023p = 0.023, while the corresponding improvement is ΔAIC=3.19\Delta\mathrm{AIC} = 3.19. This provides modest evidence that the PD increases more steeply with energy during the upper track.

Joint comparison with the intermediate state

We next jointly fit the Stokes-QQ and UU spectra from the intermediate-state observation and both SPL tracks to test directly whether the polarization evolved across the state transition. The spectral decomposition, detector gain corrections, and cross-normalization constants for each track were fixed at their corresponding Stokes-II solutions. A common POLLIN prescription was applied to the full continuum, with the PA slope fixed at zero.

The preferred model allows the upper- and lower-track polarization amplitudes and slopes to differ, while tying the intermediate-state amplitude to that of the upper track and allowing the intermediate-state slope to vary independently. A single PA was shared by all three intervals. This model gives χ2/ν=532.40/522\chi^2/\nu= 532.40/522 with a null-hypothesis probability of 0.367. The shared upper-track and intermediate-state amplitude is A1=0.2−0.2+0.7%A_1 = 0.2^{+0.7}_{-0.2}\%, with its 90% lower confidence bound reaching the hard limit at zero. The upper-track slope is 0.8±0.3% keV−10.8 \pm0.3\%\,\mathrm{keV}^{-1}, while the lower-track amplitude and slope are 2±1%2 \pm1\% and 0.1±0.6% keV−10.1 \pm0.6\%\,\mathrm{keV}^{-1}, respectively. The intermediate-state slope is 1.7−0.3+0.2% keV−11.7^{+0.2}_{-0.3}\%\,\mathrm{keV}^{-1}, and the common PA is −1±2∘-1 \pm2^\circ. All errors are reported at 90% confidence.

Tying the intermediate-state slope to the upper-track slope worsens the fit to χ2/ν=570.60/523\chi^2/\nu= 570.60/523, corresponding to Δχ2=38.20\Delta\chi^2 = 38.20 for one additional constraint and a nominal p≃6.4×10−10p \simeq6.4 \times10^{-10}. The change in the PD slope between the intermediate and fading-SPL states is therefore strongly required. In contrast, allowing the intermediate-state PA to differ improves the fit by only Δχ2=0.49\Delta\chi^2 = 0.49 for one additional parameter, while allowing its amplitude to differ improves the fit by only Δχ2=1.39\Delta\chi^2 = 1.39. Allowing both quantities to differ gives χ2/ν=530.52/520\chi^2/\nu= 530.52/520, an improvement of only Δχ2=1.88\Delta\chi^2 = 1.88 for two additional parameters. The preferred model therefore retains a common PA and ties the intermediate-state amplitude to the upper-track value while allowing an independent intermediate-state slope.

Discussion

Polarimetric evolution from the intermediate to the SPL state

The two IXPE observations provide a direct view of the polarization evolution of GS 1354–64 across a major accretion-state transition, evidenced by the disappearance of the Type-C QPO, strong suppression of rapid variability, increased prominence of the thermal disk, reduction in the broadband PD, and flattening of the PD energy dependence. During the intermediate state, the source had a 2–8 keV polarization degree of 3.9−0.2+0.3%3.9^{+0.3}_{-0.2}\% at PA=−2∘±2∘\mathrm{PA} = -2^\circ\pm2^\circ, whereas the subsequent fading-SPL observation gives PD =2.3−0.3+0.2%= 2.3^{+0.2}_{-0.3}\% at PA=4∘±3∘\mathrm{PA} = 4^\circ\pm3^\circ. Thus, the broadband PD decreased in both observed intensity tracks as the source became more disk bright, while the PA remained statistically consistent across the transition.

The state-dependent change is particularly clear in the energy dependence of the PD. The preferred joint Stokes-QQ and UU fit gives an intermediate-state slope of 1.7−0.3+0.2% keV−11.7^{+0.2}_{-0.3}\%\,\mathrm{keV}^{-1}, compared with 0.8±0.3% keV−10.8 \pm0.3\%\,\mathrm{keV}^{-1} for the upper SPL track and 0.1±0.6% keV−10.1 \pm0.6\%\,\mathrm{keV}^{-1} for the lower track. Tying the intermediate-state and upper-track slopes worsens a joint fit by Δχ2=38.20\Delta\chi^2 = 38.20 for one additional constraint, corresponding to a nominal p≃6.4×10−10p \simeq6.4 \times10^{-10}. By contrast, allowing the intermediate-state PA to differ improves the fit by only Δχ2=0.49\Delta\chi^2 = 0.49. The dominant polarimetric evolution is therefore a reduction and flattening of the PD spectrum, rather than a rotation of its PA. The consistent PAs indicate that the

Note. For the common-polarization models (Cases 1 and 2), a single polarization prescription is applied to the sum of the disk and Comptonized continua; the corresponding A1A_1 or PD is therefore shown across the two component columns. For the orthogonal-component models, the disk polarization angle was constrained to satisfy PAdisk=PAC+90∘\mathrm{PA}_{\mathrm{disk}}=\mathrm{PA}_{\mathrm{C}}+90^\circ. For POLLIN, A1,CA_{1,\mathrm{C}} is the extrapolated polarization degree of the Comptonized component at 1 keV, and Aslope,CA_{\mathrm{slope,C}} is its linear energy slope. The PA slope was fixed at zero in all POLLIN fits. For lower-track Case 3, the confidence interval on Aslope,CA_{\mathrm{slope,C}} was checked using a parameter scan because the standard error calculation reported non-monotonicity in the fit-statistic surface. The reduced χ2\chi^2 values slightly below unity for several lower-track fits (χ2/ν≃0.92\chi^2/\nu\simeq0.92–0.930.93) lie within the expected statistical dispersion for approximately 170 degrees of freedom and do not, by themselves, indicate overfitting.

A1A_1 treatmentA1A_1 (%)AslopeA_{\mathrm{slope}} treatmentAslopeA_{\mathrm{slope}} (% keV−1^{-1})PA (deg)χ2/dof\chi^2/\mathrm{dof}Δχ2\Delta\chi^2
Tied0.4−0.4+1.00.4^{+1.0}_{-0.4}Tied0.8−0.4+0.30.8^{+0.3}_{-0.4}−1−5+5-1^{+5}_{-5}342.33/345
Tied0.4−0.4+1.00.4^{+1.0}_{-0.4}Separate0.7−0.4+0.3/0.8−0.4+0.30.7^{+0.3}_{-0.4}/0.8^{+0.3}_{-0.4}−1−5+5-1^{+5}_{-5}342.02/3440.31a0.31^{a}
Separate0.2−0.4+1.0/0.8−0.7+1.00.2^{+1.0}_{-0.4}/0.8^{+1.0}_{-0.7}Tied0.8−0.4+0.20.8^{+0.2}_{-0.4}−1−5+5-1^{+5}_{-5}340.01/3442.32a2.32^{a}
Separate−0.9±1.2/2.3±1.5-0.9\pm1.2/2.3\pm1.5Separate1.2±0.5/0.1±0.61.2\pm0.5/0.1\pm0.60±50\pm5334.82/3435.19

Table 4. Simultaneous tests of track-dependent polarization using a common POLLIN prescription. Polarization parameters were tied among detector units. Values of A1A_1 and AslopeA_{\mathrm{slope}} are given in percent and percent per keV, respectively. When a parameter was allowed to differ between tracks, the values are listed as upper track/lower track. The PA was tied between tracks in all four fits, and its energy dependence was fixed to zero. All uncertainties and upper limits are reported at 90% confidence.

Note. Values separated by a slash correspond to the upper and lower tracks, respectively (except in the χ2/ν\chi^2/\nu row). In the fit with both A1A_1 and AslopeA_{\mathrm{slope}} separate, A1A_1 was allowed to take negative values because it represents an extrapolated intercept below the fitted bandpass. The negative upper-track value does not imply a physical negative PD: the modeled PD remains positive throughout 2–8 keV. The PA slope was fixed at zero in all fits. a^{a} Relative to the model in which both A1A_1 and AslopeA_{\mathrm{slope}} were tied. b^{b} Relative to the model with separate A1A_1 values and a tied AslopeA_{\mathrm{slope}}.

disk–corona system retained the same projected orientation across the transition, despite changes in their relative fluxes and the disappearance of the Type-C QPO.

Part of this evolution can plausibly arise from the changing relative fluxes of the thermal and Comptonized components. If the Comptonized emission is more strongly polarized than the directly observed disk, increasing the disk contribution lowers the band-integrated PD through dilution. The energy dependence then reflects the changing component mixture across the IXPE band, because the disk contributes most strongly at lower energies while the Comptonized component becomes relatively more important toward higher energies. This naturally produces an increasing net PD with energy. However, dilution alone is not uniquely established: changes in the intrinsic disk polarization, coronal optical depth or geometry, and scattering in the disk atmosphere could produce similar behavior. The observations therefore show that the polarization-weighted disk–corona balance changed substantially, but do not uniquely determine whether this resulted only from changing flux fractions or also from evolution of the intrinsic component polarizations.

Figure 9. Simultaneous fit to the track-resolved IXPE Stokes-QQ and UU spectra. The left and right columns show the upper and lower tracks, respectively, with marker shape identifying the three detector units. The upper panels show the unfolded Stokes spectra and the model in which both A1A_1 and AslopeA_{\mathrm{slope}} were allowed to differ between tracks, while the PA was tied. The middle panels show the residuals when the tracks have separate A1A_1 values but share a common PD slope. The lower panels show the residuals when both A1A_1 and the PD slope are track dependent. Allowing the slopes to differ improves the fit by Δχ2=5.19\Delta\chi^2 = 5.19 for one additional free parameter. This corresponds to marginal evidence that the PD increases more steeply with energy along the upper track.

The origin of the two intensity tracks

The SPL observation of GS 1354−64 reveals two distinct intensity tracks: a declining-continuum interrupted by two broad troughs, each lasting approximately 10510^5 s and a shorter-interval dip. The spectral differences between these tracks are dominated by the thermal component. In the preferred simultaneous fit, both the diskbb temperature and normalization differ between the upper and lower tracks, whereas the nthComp photon index and normalization can remain tied. Allowing the disk normalization to vary in addition to the temperature improves the fit by Δχ2=131.82\Delta\chi^2 = 131.82 for one additional free parameter, while subsequently allowing either the Compton normalization or the full Compton continuum to vary produces improvements of only Δχ2≃2\Delta\chi^2 \simeq2. The lower track spectrum is therefore not simply a fainter realization of the same continuum: it has a cooler and fainter disk with no independent change in the nthComp photon index or normalization required.

The inferred luminosity places this behavior near the regime in which BHXBs commonly begin the transition out of disk-bright states and complete the transition back toward the hard state. Assuming a distance of 7 kpc ([62]), the model-derived bolometric flux indicates that the source declined from approximately 2.5% to 1.6% of the Eddington luminosity during the observation. This overlaps the ∼1\sim1–2% Eddington range often associated with the soft-to-hard back-transition ([48, 21]). The repeated movement between the upper and lower tracks may therefore represent brief excursions along the intermediate-state branch as the source approached the transition threshold. This interpretation is supported qualitatively by both the source’s motion in the IXPE HID and its behavior in the PD–PA plane, both indicating a possible sputtering near the cusp of a state transition, as the source repeatedly moves between neighboring configurations before settling into the hard state again.

This behavior differs from that expected for a conventional cold-absorption dip. In wind-fed systems such as Cygnus X-1, dips preferentially suppress the soft X-ray emission, produce spectral hardening, and are commonly described by changes in neutral or partially covering absorption. While GS 1354−64 is putatively a Roche-lobe-overflow low-mass X-ray binary rather than a wind-fed system ([12]), we tested this possibility directly by tying the intrinsic continuum between the two GS 1354−64 tracks while allowing NHN_{\mathrm{H}} to differ. Although the trough favored a larger nominal column density, the resulting fit remained grossly unacceptable, with χ2/ν=28403.9/166\chi^2/\nu= 28403.9/166, and drove the tied continuum to extreme compensating values. Variable neutral absorption alone therefore cannot reproduce the track differences. Recent IXPE analysis of dipping in Cygnus X-1 found that the broadband PD and PA remained approximately unchanged between dip and nondip intervals, consistent with compact absorbing structures crossing the line of sight without substantially changing the geometry of the extended polarized source ([43]). The GS 1354−64 troughs are broader and smoother and are instead described successfully by changes in the thermal emission.

More complex obscuration cannot yet be excluded completely. Although the absorption-only Case 0 rules out a change in fully covering neutral absorption as the sole origin of the two tracks, the limited spectral resolution and 2–8 keV bandpass of IXPE provide weaker constraints on partially covering or highly ionized material. A highly ionized or nearly gray scattering screen could reduce the observed intensity without producing strong hardness changes though would not explain the suggestive temperature change across tracks. Nevertheless, maintaining the source in a low state for ∼105\sim10^5 s would require a large and coherent line-of-sight structure. As a simple order-of-magnitude estimate, following the crossing-time argument commonly applied to X-ray occultation events (e.g., [74]), consider an obscuring structure moving across the line of sight with an assumed transverse velocity v⊥v_{\perp}. To maintain the source in the trough for ttrought_{\mathrm{trough}}, its projected extent along the direction of motion must satisfy

D⊥≳v⊥ttrough≃5×1012(v⊥500 km s−1)(ttrough105 s) cm.(2)D_{\perp} \gtrsim v_{\perp}t_{\mathrm{trough}} \simeq5 \times10^{12} \left(\frac{v_{\perp}}{500\ \mathrm{km\ s}^{-1}}\right)\left(\frac{t_{\mathrm{trough}}}{10^5\ \mathrm{s}}\right)\ \mathrm{cm}. \tag*{(2)}

Here, 500 km s−1^{-1} is an illustrative velocity rather than a measured value for GS 1354−64. This is many orders of magnitude larger than the characteristic X-ray-emitting region and may approach or exceed the scale of the binary. Since the ingress and egress are substantially shorter than the trough duration (∼\sim5–11 minutes, compared with the ∼105\sim10^5-s or ∼\sim28-hr trough duration), the obscurer would additionally require a sharp leading edge and a large transverse aspect ratio. For plausible binary-scale velocities, this would require a substantially extended and coherent obscuring structure, unlike the compact clouds inferred from rapid X-ray occultation events in Active Galactic Nuclei (AGN) such as NGC 1365 ([74]). The measured rapid egress strengthens this geometric constraint, while broader-band fits using partial-covering or ionized-absorption models would be required to test these more complex obscuration scenarios.

The fitted DISKBB normalization should not, however, be interpreted directly as a change in the physical inner radius. In addition to its dependence on the assumed distance, inclination, color-correction factor, inner-boundary condition, and any partial occultation, the normalization may be biased by the treatment of Comptonization. In the additive DISKBB+NTHCOMP model used here, the seed-photon temperature is tied to the disk temperature, but the Comptonized flux is not coupled self-consistently to the number of disk photons removed by scattering. The fitted DISKBB component therefore represents only the directly observed disk emission; any disk photons redistributed into the Comptonized component appear as “missing” disk flux, biasing the inferred radius low. A photon-conserving convolution treatment such as SIMPL would be required for a more physical radius estimate ([82]); however, at the high electron temperatures inferred here, a Comptonization model employing an accurate Klein–Nishina cross-section, such as the plane-parallel COMPPS or COMPP S_CONV model, would be more appropriate (J. Poutanen & R. Svensson [64]). Likewise, the inclination inferred from the phenomenological LAOR component should not be treated as a robust measurement of the binary or inner-disk inclination. It is sensitive to the assumed line energy, emissivity profile, disk radii, continuum model, and instrumental calibration, several of which were fixed in these fits. The combination of a lower temperature and smaller apparent normalization during the lower track could therefore represent a genuine change in the visible thermal-emitting area, a change in the disk atmosphere or spectral hardening, or obscuration of part of an extended disk. The current data establish that the phenomenological change is associated primarily with the thermal component, but do not uniquely distinguish among these physical explanations.

Polarimetric constraints on the trough geometry

The upper track favors a PD that increases with energy, whereas the lower track is adequately described by energy-independent polarization. The direct simultaneous comparison provides only modest evidence for different slopes, with the inferred difference sensitive to the covariance between the extrapolated POLLIN intercept and slope. The track dependence should therefore be regarded as suggestive rather than a firm detection of polarimetric evolution during the troughs.

The component-resolved fits illustrate an important degeneracy. The rising net PD along the upper track can be reproduced either by intrinsic energy dependence or by a weakly polarized disk diluting a more strongly polarized Comptonized component. Models with an unpolarized disk yield similar Compton polarization degrees in both tracks, while models with orthogonally polarized components are also acceptable but poorly constrain their individual contributions. The data therefore do not uniquely distinguish changes in intrinsic polarization or the scattering environment from changes in component weighting.

The stable PA provides a complementary constraint: neither track requires a change in the net polarization orientation. This disfavors explanations requiring a large rotation of the observed polarization during the troughs, but does not imply that the accretion geometry is otherwise unchanged. Changes in the visible disk area, temperature, or scattering properties remain compatible with the measurements.

The possible track dependence is weaker than the polarization evolution across the intermediate-to-SPL transition, where the PD spectrum becomes significantly shallower without a measurable PA change. Importantly, a shallower PD spectrum need not indicate a weaker corona. If the energy dependence arises partly from disk dilution, the reduced thermal contribution during the troughs can allow polarized Comptonized emission to dominate over more of the IXPE band, flattening the net PD spectrum without requiring a change in its intrinsic polarization.

A changing-state analogy across the black-hole mass scale

The repeated movement of GS 1354−64 between two long-lived intensity tracks invites comparison with changing-state accretion behavior. Changing-state AGN (CSAGN) such as Mrk 590 and Mrk 1018 undergo large continuum and broad-line changes that are increasingly interpreted as changes in the accretion flow rather than solely as variable obscuration ([73]). Mrk 1018, for example, evolved between bright and faint states over years, accompanied by substantial UV and X-ray changes and evidence for a reconfiguration of the inner accretion flow ([91]). Mrk 590 has likewise displayed fading, rebrightening, and repeated X-ray/UV activity at relatively low Eddington ratio ([41]). These sources demonstrate that accretion flows can occupy and revisit distinct luminosity configurations without a simple monotonic evolution.

An additional relevant comparison is the “flip-flop” behavior observed in some BHXBs near global state transitions ([11, 47]). These events involve rapid transitions between two distinct intensity and spectral-timing configurations, each of which can persist substantially longer than the transition itself. This morphology resembles the repeated upper- and lower-track behavior of GS 1354−64: the best-sampled transition, the egress from the second broad trough, has a measured 10–90% duration of approximately 5–11 minutes, whereas the trough itself persists for ∼105\sim10^{5} s.

T. Maccarone et al. [47] showed that the ingress, egress, and state-duration timescales of XRB flip-flops and CSAGN are broadly consistent with scaling linearly with black-hole mass. Under this scaling,

tAGN≃1.9(tXRB10 min)(MAGN/MXRB105) yr,(3)t_{\mathrm{AGN}} \simeq1.9 \left(\frac{t_{\mathrm{XRB}}}{10\ \mathrm{min}}\right) \left(\frac{M_{\mathrm{AGN}}/M_{\mathrm{XRB}}}{10^{5}}\right)\ \mathrm{yr}, \tag*{(3)}

so the measured 5–11 minute egress would correspond to approximately 1.0–2.1 yr for a 106 M⊙10^{6}\ M_{\odot} AGN and a 10 M⊙10\ M_{\odot} XRB. This scaled timescale lies within the months-to-years range observed in CSAGN such as 1ES 1927+654 ([88]) and SDSS J162829.17+432948.5 ([95]), although this agreement remains phenomenological and does not establish a common physical mechanism.

The luminosity at which the transitions occur provides a complementary comparison. Both XRB flip-flops and changing-state transitions preferentially occur at luminosities of a few percent of Eddington [59, 47]. Extrapolating the best-fitting time-averaged IXPE spectral model over 0.01–100 keV and removing the foreground absorption gives an unabsorbed model flux of Fbol=4.19×10−9 erg cm−2 s−1F_{\mathrm{bol}} = 4.19 \times10^{-9}\ \mathrm{erg\ cm^{-2}\ s^{-1}}, although this bolometric correction is model dependent. For a 10 M⊙10\ M_{\odot} black hole,

LbolLEdd≃1.95×10−2(Fbol4.19×10−9 erg cm−2 s−1)×(d7 kpc)2(M10 M⊙)−1.(4)\frac{L_{\mathrm{bol}}}{L_{\mathrm{Edd}}} \simeq1.95 \times10^{-2} \left(\frac{F_{\mathrm{bol}}}{4.19 \times10^{-9}\ \mathrm{erg\ cm^{-2}\ s^{-1}}}\right) \times\left(\frac{d}{7\ \mathrm{kpc}}\right)^2 \left(\frac{M}{10\ M_{\odot}}\right)^{-1}. \tag*{(4)}

The large uncertainty in the distance prevents this estimate from locating GS 1354−64 securely relative to the putative 1% transition scale. At the nominal distance of 7 kpc, the inferred Eddington ratio is approximately 2%, consistent with a source in the SPL state approaching a return to the hard state, since BHXBs commonly complete the soft-to-hard transition at a few percent of Eddington [48, 21]. By contrast, distances of 0.7 and 25 kpc would imply Lbol/LEdd≃2×10−4L_{\mathrm{bol}}/L_{\mathrm{Edd}} \simeq2 \times10^{-4} and 0.25, respectively.

Despite these similarities in morphology, transition timescale, and inferred luminosity, GS 1354−64 does not reproduce every defining property of previously reported flip-flops. Classical examples often show marked transitions between QPO-dominated and broadband-noise-dominated variability, whereas both tracks in GS 1354−64 retain weak variability without a detected QPO. The present observations may therefore represent a less complete excursion between neighboring accretion configurations near the soft-to-hard transition, rather than a direct example of the classical flip-flop phenomenon.

Conclusion

The IXPE observations of GS 1354−64 reveal evolution on two distinct levels. Between the preceding intermediate-state observation and the fading SPL observation, the disappearance of the Type-C QPO, strong suppression of rapid variability, increased prominence of the thermal disk, and flattening of the PD energy dependence indicate a substantial reconfiguration of the emitting and scattering flow. The persistence of a steep Comptonized tail argues against identifying the later observation as a purely thermal soft state, while the stable PA suggests that the evolution occurred without a measurable reorientation of the projected symmetry axis.

Within the fading SPL state, the recurrent tracks are distinguished robustly by changes in the thermal component, while their polarization behavior is suggestive of a pivoting accretion power dissipation in the disk-vs-corona or changes in the optical depth or geometry of the material scattering disk photons. This interpretation nevertheless remains dependent on the adopted polarization decomposition. The marginal difference between the upper-track and lower-track PD slopes, together with the significantly steeper intermediate-state slope, indicates that the disk–corona weighting may change among the different luminosity configurations. In this sense, GS 1354−64 may represent a stellar-mass example of an accretion flow moving between recurrent luminosity configurations, with a phenomenological connection to changing-state behavior across the black-hole mass scale.

IXPE Timing Analysis

We analyze the IXPE timing data with version 1.1.2.4 of the Python package Stingray [30, 7], following the procedure outlined by S. Ravi et al. [69]. After filtering the good-time intervals from the IXPE data, we combine events from the three DU's into light curves with 1 ms time resolution, corresponding to a Nyquist frequency of 500 Hz. We compute the power density spectrum by averaging 100 s intervals and adopt the fractional-rms normalization. We correct for deadtime using the Fourier Amplitude Difference method [6], then logarithmically rebin the spectrum using bins with 5% fractional width. The Poisson-noise level is estimated from the average power above 100 Hz and then subtracted.

The deadtime-corrected, logarithmically rebinned power density spectra (PDS) extracted from the full IXPE SPL observation and separately from the upper and lower intensity tracks identified in the light curve (Figure 1) are shown in Figure 10. In contrast to the intermediate-state observation presented in S. Ravi et al. [69], the SPL PDS exhibits characteristically weaker broadband variability and no evidence for QPOs. We examine both the full observation and the individual intensity tracks to track the variability across the SPL state.

Figure 10. Noise-subtracted PDS plotted as ν[P(ν)−Pnoise]\nu[P(\nu)-P_{\mathrm{noise}}] in units of fractional rms2\mathrm{rms}^2 for the full GS 1354−64 IXPE SPL observation (top), the upper track (middle), and the lower track (bottom).

We first fit the full SPL PDS in XSPEC using a constant continuum and a zero-centered Lorentzian to describe broadband variability, obtaining an acceptable fit with χ2/ν=171.01/154\chi^2/\nu= 171.01/154. Adding further Lorentzian components did not significantly improve the fit, and no additional broadband variability or QPO component was required by the SPL PDS. We confirm that the previously detected oscillation is not significantly recovered in the full SPL observation. This nondetection, the collapsed broadband noise, and integrated 0.1–10 Hz rms power of 3.5%–5.3%, is consistent with softer states (e.g., P. Casella et al. [14]; T. M. Belloni [9]; A. R. Ingram & S. E. Motta 2019).

We additionally fit the upper- and lower-track PDS using a power-law continuum and tested for additional broadband variability by adding a zero-centered Lorentzian. The measured noise-subtracted 0.1–10 Hz powers are 1.25×10−31.25 \times10^{-3} and 2.82×10−3 rms22.82 \times10^{-3}\,\mathrm{rms}^2 for the upper and lower tracks, respectively, corresponding to fractional rms amplitudes of 3.54±1.29%3.54 \pm1.29\% and 5.31±1.12%5.31 \pm1.12\%. The added Lorentzian therefore represents only weak broadband variability in both tracks, rather than evidence for a discrete QPO.

Acknowledgments

This work reports observations obtained with the Imaging X-ray Polarimetry Explorer (IXPE), a joint US (NASA) and Italian (ASI) mission, led by Marshall Space Flight Center (MSFC). The research uses data products provided by the IXPE Science Operations Center (MSFC), using algorithms developed by the IXPE Collaboration (MSFC, Istituto Nazionale di Astrofisica - INAF, Istituto Nazionale di Fisica Nucleare - INFN, ASI Space Science Data Center - SSDC), and distributed by the High-Energy Astrophysics Science Archive Research Center (HEASARC). This research has made use of data from the NuSTAR mission, a project led by the California Institute of Technology, managed by the Jet Propulsion Laboratory, and funded by the National Aeronautics and Space Administration. Data analysis was performed using the NuSTAR Data Analysis Software (NuSTARDAS), jointly developed by the ASI Science Data Center (SSDC, Italy) and the California Institute of Technology (USA). This research has made use of MAXI data provided by RIKEN, JAXA and the MAXI team. We thank Andrzej Zdziarski for helpful comments on this source.

SR acknowledges support from NASA through the Imaging X-ray Polarimetry Explorer (IXPE) General Observer Program under grant No. 80NSSC26K1006, supporting the project “Energy-dependent Polarization Rotation: Testing General Relativistic Disk Models with IXPE.” JFS was supported by the Chandra X-ray Observatory, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-03060. AV acknowledges support from the Academy of Finland grants 355672 and 372881. AI is supported by the European Union (ERC, X-MAPS, 101169908). M.D., J.P., and J.S. acknowledge support from the Czech Science Foundation grant 26-22614S.

Facilities: IXPE [93, 79], MAXI [51], NuSTAR [28]

Software: astropy [5, 3, 4], NumPy [27], SciPy [92], IPython [61], bilby [2], dynesty [80], ixpeobssim [8], QUEEN-BEE [67, 68], Stingray [30, 7], SAOImage DS9 [33], XSPEC [1]

References

  1. [1]Arnaud, K. A. 1996, in Astronomical Society of the Pacific Conference Series, Vol. 101, Astronomical Data Analysis Software and Systems V, ed. G. H. Jacoby & J. Barnes, 17
  2. [2]Ashton, G., Hübner, M., Lasky, P. D., et al. 2019, ApJS, 241, 27, doi: 10.3847/1538-4365/ab06fcDOI
  3. [3]Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f
  4. [4]Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74DOI
  5. [5]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068
  6. [6]Bachetti, M., & Huppenkothen, D. 2018, ApJL, 853, L21, doi: 10.3847/2041-8213/aaa83bDOI
  7. [7]Bachetti, M., Huppenkothen, D., Khan, U., et al. 2023,, v1.1.2 Zenodo, doi: 10.5281/zenodo.7970570
  8. [8]Baldini, L., Bucciantini, N., Lalla, N. D., et al. 2022, SoftwareX, 19, 101194, doi: 10.1016/j.softx.2022.101194DOI
  9. [9]Belloni, T. M. 2010, in Lecture Notes in Physics, Berlin Springer Verlag, ed. T. Belloni, Vol. 794, 53, doi: 10.1007/978-3-540-76937-8_3
  10. [11]Buisson, D. J. K., Marcel, G., López-Barquero, V., et al. 2025, arXiv e-prints, arXiv:2502.08718, doi: 10.48550/arXiv.2502.08718DOI
  11. [12]Casares, J., Zurita, C., Shahbaz, T., Charles, P. A., & Fender, R. P. 2004, ApJL, 613, L133, doi: 10.1086/425145DOI
  12. [13]Casares, J., Orosz, J. A., Zurita, C., et al. 2009, ApJS, 181, 238, doi: 10.1088/0067-0049/181/1/238DOI
  13. [14]Casella, P., Belloni, T., & Stella, L. 2005, ApJ, 629, 403, doi: 10.1086/431174DOI
  14. [15]Chandrasekhar, S. 1960, Radiative transfer
  15. [16]Connors, P. A., & Stark, R. F. 1977, Nature, 269, 128, doi: 10.1038/269128a0DOI
  16. [17]Di Marco, A., Soffitta, P., Costa, E., et al. 2023, AJ, 165, 143, doi: 10.3847/1538-3881/acab0f
  17. [18]Done, C., Gierliński, M., & Kubota, A. 2007, A&A Rv, 15, 1, doi: 10.1007/s00159-007-0006-1DOI
  18. [19]Dovčiak, M., Muleri, F., Goosmann, R. W., Karas, V., & Matt, G. 2008, MNRAS, 391, 32, doi: 10.1111/j.1365-2966.2008.13872.xDOI
  19. [20]Draghis, P. A., Miller, J. M., Costantini, E., et al. 2024, ApJ, 969, 40, doi: 10.3847/1538-4357/ad43eaDOI
  20. [21]Dunn, R. J. H., Fender, R. P., Körding, E. G., Belloni, T., & Cabanac, C. 2010, MNRAS, 403, 61, doi: 10.1111/j.1365-2966.2010.16114.xDOI
  21. [22]Ebisawa, K. 1991, PhD thesis, JAXA, Institute of Space and Astronautical Science
  22. [23]El-Batal, A. M., Miller, J. M., Reynolds, M. T., et al. 2016, ApJL, 826, L12, doi: 10.3847/2041-8205/826/1/L12
  23. [24]Ewing, M., Parra, M., Mastrosserio, G., et al. 2025, MNRAS, 541, 1774, doi: 10.1093/mnras/staf859
  24. [25]Fender, R. P., Belloni, T. M., & Gallo, E. 2004, MNRAS, 355, 1105, doi: 10.1111/j.1365-2966.2004.08384.xDOI
  25. [26]Gandhi, P., Rao, A., Johnson, M. A. C., Paice, J. A., & Maccarone, T. J. 2019, MNRAS, 485, 2642, doi: 10.1093/mnras/stz438DOI
  26. [27]Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2DOI
  27. [28]Harrison, F. A., Craig, W. W., Christensen, F. E., et al. 2013, ApJ, 770, 103, doi: 10.1088/0004-637X/770/2/103DOI
  28. [29]Homan, J., & Belloni, T. 2005, Ap&SS, 300, 107, doi: 10.1007/s10509-005-1197-4DOI
  29. [30]Huppenkothen, D., Bachetti, M., Stevens, A. L., et al. 2019, apj, 881, 39, doi: 10.3847/1538-4357/ab258dDOI
  30. [31]Ingram, A., Belomestjev, N., Veledina, A., et al. 2024, ApJ, 968, 76, doi: 10.3847/1538-4357/ad3faf Ingram, A. R., & Motta, S. E. 2019, NewAR, 85, 101524, doi: 10.1016/j.newar.2020.101524DOI
  31. [32]Jeffreys, H. 1998, Theory of Probability, 3rd edn. (Oxford University Press)
  32. [33]Joye, W. A., & Mandel, E. 2003, in Astronomical Society of the Pacific Conference Series, Vol. 295, Astronomical Data Analysis Software and Systems XII, ed. H. E. Payne, R. I. Jedrzejewski, & R. N. Hook, 489
  33. [34]Kaastra, J. S., & Bleeker, J. A. M. 2016, A&A, 587, A151, doi: 10.1051/0004-6361/201527395DOI
  34. [35]Kislat, F., Clark, B., Beilicke, M., & Krawczynski, H. 2015, Astroparticle Physics, 68, 45, doi: 10.1016/j.astropartphys.2015.02.007
  35. [36]Kitamoto, S., Tsunemi, H., Pedersen, H., Ilovaisky, S. A., & van der Klis, M. 1990, ApJ, 361, 590, doi: 10.1086/169222DOI
  36. [37]Koljonen, K. I. I., Russell, D. M., Corral-Santana, J. M., et al. 2016, MNRAS, 460, 942, doi: 10.1093/mnras/stw1007DOI
  37. [38]Kravtsov, V., Bocharova, A., Veledina, A., et al. 2025, A&A, 701, A115, doi: 10.1051/0004-6361/202555411DOI
  38. [39]Krawczynski, H., Muleri, F., Dovčiak, M., et al. 2022, Science, 378, 650, doi: 10.1126/science.add5399DOI
  39. [40]Laor, A. 1991, ApJ, 376, 90, doi: 10.1086/170257DOI
  40. [41]Lawther, D., Vestergaard, M., Raimundo, S., et al. 2023, MNRAS, 519, 3903, doi: 10.1093/mnras/stac3515DOI
  41. [42]Lightman, A. P., & Shapiro, S. L. 1975, ApJL, 198, L73, doi: 10.1086/181815DOI
  42. [43]Ling, Y.-s., Xie, F., Di Marco, A., La Monaca, F., & Ge, M.-Y. 2026, arXiv e-prints, arXiv:2607.26144, doi: 10.48550/arXiv.2607.26144DOI
  43. [44]Liu, H., Jiang, J., Zhang, Z., et al. 2023, ApJ, 951, 145, doi: 10.3847/1538-4357/acd8b9DOI
  44. [45]Liu, H., Kong, L., Adeogoke, O. K., et al. 2026, arXiv e-prints, arXiv:2603.06883, doi: 10.48550/arXiv.2603.06883DOI
  45. [46]Loskutov, V. M., & Sobolev, V. V. 1979, Astrofizika, 15, 241
  46. [47]Maccarone, T., Runnoe, J. C., Marcel, G., et al. 2026, PASA, 43, e044, doi: 10.1017/pasa.2026.10176DOI
  47. [48]Maccarone, T. 2003, A&A, 409, 697, doi: 10.1051/0004-6361:20031146DOI
  48. [49]Marra, L., Brigitte, M., Rodriguez Cavero, N., et al. 2024, A&A, 684, A95, doi: 10.1051/0004-6361/202348277DOI
  49. [50]Mastroserio, G., De Marco, B., Baglio, M. C., et al. 2025, ApJL, 978, L19, doi: 10.3847/2041-8213/ad9913DOI
  50. [51]Matsuoka, M., Kawasaki, K., Ueno, S., et al. 2009, PASJ, 61, 999, doi: 10.1093/pasj/61.5.999DOI
  51. [52]McClintock, J. E., Narayan, R., & Steiner, J. F. 2014, SSRv, 183, 295, doi: 10.1007/s11214-013-0003-9DOI
  52. [53]McClintock, J. E., & Remillard, R. A. 2006, in Compact stellar X-ray sources, ed. W. H. G. Lewin & M. van der Klis, Vol. 39, 157–213, doi: 10.48550/arXiv.astro-ph/0306213
  53. [54]Mikusincova, R., Dovciak, M., Bursa, M., et al. 2023, MNRAS, 519, 6138, doi: 10.1093/mnras/stad077DOI
  54. [55]Mitsuda, K., Inoue, H., Koyama, K., et al. 1984, PASJ, 36, 741
  55. [56]Negoro, H., Fujiwara, K., Nakajima, M., et al. 2025, The Astronomer’s Telegram, 17563, 1
  56. [57]Negoro, H., Nakajima, M., Takagi, K., et al. 2026, The Astronomer’s Telegram, 17618, 1
  57. [58]Niedźwiecki, A., Szanecki, M., & Zdziarski, A. A. 2019, MNRAS, 485, 2942, doi: 10.1093/mnras/stz487DOI
  58. [59]Noda, H., & Done, C. 2018, MNRAS, 480, 3898, doi: 10.1093/mnras/sty2032DOI
  59. [60]Pahari, M., Gandhi, P., Charles, P. A., et al. 2017, MNRAS, 469, 193, doi: 10.1093/mnras/stx840DOI
  60. [61]Pérez, F., & Granger, B. E. 2007, Computing in Science and Engineering, 9, 21, doi: 10.1109/MCSE.2007.53
  61. [62]Plotkin, R. M., Bahramian, A., Miller-Jones, J. C. A., et al. 2021, MNRAS, 503, 3784, doi: 10.1093/mnras/stab644DOI
  62. [63]Podgorný, J., Svoboda, J., Dovčiak, M., et al. 2024, A&A, 686, L12, doi: 10.1051/0004-6361/202450566DOI
  63. [64]Poutanen, J., & Svensson, R. 1996, ApJ, 470, 249, doi: 10.1086/177865DOI
  64. [65]Poutanen, J., & Vurm, I. 2009, ApJL, 690, L97, doi: 10.1088/0004-637X/690/2/L97DOI
  65. [66]Ratheesh, A., Dovčiak, M., Krawczynski, H., et al. 2024, ApJ, 964, 77, doi: 10.3847/1538-4357/ad226eDOI
  66. [67]Ravi, S., Ng, M., & Marshall, H. L. 2025, v3.2 Zenodo, doi: 10.5281/zenodo.17068831DOI
  67. [68]Ravi, S., Ng, M., Marshall, H. L., & Gnarini, A. 2026a, ApJ, 997, 60, doi: 10.3847/1538-4357/ae21bdDOI
  68. [69]Ravi, S., Marra, L., Steiner, J. F., et al. 2026b, arXiv e-prints, arXiv:2603.03463, doi: 10.48550/arXiv.2603.03463DOI
  69. [70]Rees, M. J. 1975, MNRAS, 171, 457, doi: 10.1093/mnras/171.3.457DOI
  70. [71]Remillard, R. A., & McClintock, J. E. 2006, ARA&A, 44, 49, doi: 10.1146/annurev.astro.44.051905.092532DOI
  71. [72]Reynolds, C. S. 2021, ARA&A, 59, 117, doi: 10.1146/annurev-astro-112420-035022DOI
  72. [73]Ricci, C., & Trakhtenbrot, B. 2023, Nature Astronomy, 7, 1282, doi: 10.1038/s41550-023-02108-4
  73. [74]Risaliti, G., Elvis, M., Fabbiano, G., et al. 2007, ApJL, 659, L111, doi: 10.1086/517884DOI
  74. [75]Rodriguez Cavero, N., Marra, L., Krawczynski, H., et al. 2023, ApJL, 958, L8, doi: 10.3847/2041-8213/acfd2cDOI
  75. [76]Schnittman, J. D., & Krolik, J. H. 2009, ApJ, 701, 1175, doi: 10.1088/0004-637X/701/2/1175DOI
  76. [77]Schnittman, J. D., & Krolik, J. H. 2010, ApJ, 712, 908, doi: 10.1088/0004-637X/712/2/908DOI
  77. [78]Sobolev, V. V. 1963, A treatise on radiative transfer.
  78. [79]Soffitta, P., Baldini, L., Bellazzini, R., et al. 2021, AJ, 162, 208, doi: 10.3847/1538-3881/ac19b0DOI
  79. [80]Speagle, J. S. 2020, MNRAS, 493, 3132, doi: 10.1093/mnras/staa278DOI
  80. [81]Stark, R. F., & Connors, P. A. 1977, Nature, 266, 429, doi: 10.1038/266429a0DOI
  81. [82]Steiner, J. F., Narayan, R., McClintock, J. E., & Ebisawa, K. 2009, PASP, 121, 1279, doi: 10.1086/648535DOI
  82. [83]Steiner, J. F., Nathan, E., Hu, K., et al. 2024, ApJL, 969, L30, doi: 10.3847/2041-8213/ad58e4DOI
  83. [84]Stiele, H., & Kong, A. K. H. 2016, MNRAS, 459, 4038, doi: 10.1093/mnras/stw903DOI
  84. [85]Svoboda, J., Dovčiak, M., Steiner, J. F., et al. 2024a, ApJ, 960, 3, doi: 10.3847/1538-4357/ad0842DOI
  85. [86]Svoboda, J., Dovčiak, M., Steiner, J. F., et al. 2024b, ApJL, 966, L35, doi: 10.3847/2041-8213/ad402eDOI
  86. [87]Taverna, R., Marra, L., Bianchi, S., et al. 2021, MNRAS, 501, 3393, doi: 10.1093/mnras/staa3859DOI
  87. [88]Trakhtenbrot, B., Arcavi, I., MacLeod, C. L., et al. 2019, ApJ, 883, 94, doi: 10.3847/1538-4357/ab39e4DOI
  88. [89]Veledina, A., Muleri, F., Dovčiak, M., et al. 2023, ApJL, 958, L16, doi: 10.3847/2041-8213/ad0781DOI
  89. [90]Verner, D. A., Ferland, G. J., Korista, K. T., & Yakovlev, D. G. 1996, ApJ, 465, 487, doi: 10.1086/177435DOI
  90. [91]Veronese, S., Vignali, C., Severgnini, P., Matzeu, G. A., & Cignoni, M. 2024, A&A, 683, A131, doi: 10.1051/0004-6361/202348098DOI
  91. [92]Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2DOI
  92. [93]Weisskopf, M. C., Soffitta, P., Baldini, L., et al. 2022, Journal of Astronomical Telescopes, Instruments, and Systems, 8, 026002, doi: 10.1117/1.JATIS.8.2.026002DOI
  93. [94]Wilms, J., Allen, A., & McCray, R. 2000, ApJ, 542, 914, doi: 10.1086/317016DOI
  94. [95]Zeltyn, G., Trakhtenbrot, B., Eracleous, M., et al. 2022, ApJL, 939, L16, doi: 10.3847/2041-8213/aca947
  95. [96]Życki, P. T., Done, C., & Smith, D. A. 1999, MNRAS, 309, 561, doi: 10.1046/j.1365-8711.1999.02885.xDOI

Paper details

vlnc:W_zsd5whxv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 2

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents