GX 349+2: Sco-like Z Source in NS-LMXBs

• GX 349+2 is a bright Sco-like Z source neutron star low-mass X-ray binary accreting near the Eddington limit, characterized by distinct normal and flaring branches with occasional horizontal branch intervals.
• Multi-mission studies using RXTE, NuSTAR, IXPE, and XMM-Newton have resolved its spectral components—disk, boundary layer, Comptonization, and Fe-K reflection—and documented branch-dependent timing lags ranging from seconds to hundreds of seconds.
• Integrated analyses demonstrate that GX 349+2 serves as a key laboratory for probing accretion disk geometry, slow viscous adjustments in the boundary layer, and the interplay of disk, corona, and reflection in neutron star systems.

Searching arXiv for recent GX 349+2 papers to ground the article. GX 349+2, also known as Sco X-2, is a bright Sco-like Z source in the class of neutron-star low-mass X-ray binaries (NS-LMXBs) accreting at near-Eddington luminosities. In hardness–intensity and color–color diagrams it traces predominantly the normal branch (NB) and flaring branch (FB), with a weak or absent horizontal branch (HB) in many observations, although HB intervals have been identified in at least one XMM-Newton study (Ding et al., 2015, R. et al., 5 Feb 2026). Across recent work, GX 349+2 has emerged as a key laboratory for studying the coupling between the inner accretion disk, the neutron-star boundary or spreading layer, Comptonizing plasma, relativistic reflection, long and short X-ray lags, and weak but measurable X-ray polarization. These observables together indicate a near-surface accretion flow in which branch-dependent geometric and radiative changes occur on timescales from seconds to hundreds of seconds, while the source remains persistently luminous and spectrally complex (Coughenour et al., 2017, Kashyap et al., 1 May 2025, R. et al., 5 Feb 2026).

1. Source class, nomenclature, and phenomenology

GX 349+2 is classified as a Sco-like Z source, that is, a weakly magnetized NS-LMXB accreting at near-Eddington rates (Ding et al., 2015, Kumar et al., 6 Apr 2025). In this class, motion along the Z track is conventionally described in terms of the HB, NB, and FB, though GX 349+2 often shows a long NB and elongated FB with little or no HB in many campaigns (Ding et al., 2015, Coughenour et al., 2017). The source is therefore frequently compared with Sco X-1 rather than with Cyg-like Z sources.

Several studies adopt a distance of 9.2 kpc for converting fluxes to luminosities and emitting radii (R. et al., 5 Feb 2026, Kashyap et al., 1 May 2025, Monaca et al., 9 Jul 2025). At this distance, the observed X-ray fluxes place GX 349+2 at luminosities of order $10^{38}\,\mathrm{erg\,s^{-1}}$, consistent with its identification as a near-Eddington system (Kashyap et al., 1 May 2025). NuSTAR observations covering the full Z track inferred Eddington fractions rising from $\sim 0.70$ in the NB or vertex to $\sim 1.41$ in the brightest FB state, using 3–30 keV fluxes and a canonical $1.4\,M_\odot$ neutron star (Coughenour et al., 2017).

The source has long been known to exhibit strong flaring activity, broad iron-line emission, and characteristic Z-source timing behavior. Recent work extends that picture by connecting branch-resolved timing, reflection, and polarimetry to the structure of the innermost accretion flow. This suggests that GX 349+2 is especially informative for testing how the disk, boundary layer, corona, and reflection region reorganize as the source moves along the Z track (Coughenour et al., 2017, Gnarini et al., 30 Jun 2025, R. et al., 5 Feb 2026).

2. Z-track morphology and observational campaigns

A systematic RXTE study analyzed 138 observations from 1996 to 2011 and found that the most extensive Z track was traced during 1998 January 9–29 (Ding et al., 2015). In that epoch, the hardness–intensity diagram was partitioned into 23 regions, with 8 on the NB and 15 on the FB; no HB was observed (Ding et al., 2015). The HID used hardness $=$ count-rate ratio $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$ and intensity $=$ count rate in $2.0\text{–}19.7~\mathrm{keV}$ (Ding et al., 2015). The source executed repeated FB-to-NB cycles during that interval.

A two-day NuSTAR observation beginning 2016-06-06 (ObsID 30201026002) covered approximately the full Z track and selected five spectral regions: NB, the NB–FB vertex (VX, the soft apex), FB1, FB2, and FB3 (Coughenour et al., 2017). Hardness was defined as $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$, and no HB was detected, consistent with the Sco-like classification (Coughenour et al., 2017). This campaign was central in establishing that the broad Fe K$\alpha$ line persists across states.

An XMM-Newton EPIC-pn Timing-mode observation on 2008-03-19 (ObsID 0506110101) provided a different view by isolating eleven $\sim 0.70$0 s timing segments and three $\sim 0.70$1 s transition windows, using soft and hard bands of $\sim 0.70$2 keV and $\sim 0.70$3 keV, respectively (R. et al., 5 Feb 2026). In that dataset, two segments in the HID locus were identified as HB, while the remaining segments were assigned to NB or FB. This is significant because it enabled a direct comparison of cross-correlation morphology among all three canonical branches within a single observation (R. et al., 5 Feb 2026).

The 2024 IXPE campaign, combined with NuSTAR and in some analyses Swift/XRT, added a polarization-based state diagnostic. One model-independent analysis found GX 349+2 mainly in the NB with brief FB excursions and no HB during the IXPE observation (Gnarini et al., 30 Jun 2025). Another IXPE+NuSTAR analysis described the source as predominantly in NB during the first NuSTAR pointing and mainly in NB with short excursions into the soft apex (SA) during the second, with no deep FB coverage (Monaca et al., 9 Jul 2025). These differences reflect distinct state-selection strategies and are relevant when comparing branch-resolved polarization claims.

3. Spectral structure: disk, boundary layer, Comptonization, and iron-line reflection

Broadband and medium-band spectroscopy consistently require multiple continuum components. In IXPE+NuSTAR spectro-polarimetric modeling, the continuum was represented by an absorbed combination of disk blackbody, hotter blackbody, and Comptonization, plus a line or reflection component (Kashyap et al., 1 May 2025). Specifically, one study used

$\sim 0.70$4

interpreting diskbb as the disk, bbodyrad as the neutron-star surface or boundary/spreading layer, nthcomp as thermal Comptonization, and diskline as the Fe-K reflection feature (Kashyap et al., 1 May 2025). Branch-resolved fits gave $\sim 0.70$5 keV (SA), $\sim 0.70$6 keV (FB), and $\sim 0.70$7 keV (NB), while the hotter blackbody remained near $\sim 0.70$8 keV, with an emitting radius $\sim 0.70$9 km at 9.2 kpc (Kashyap et al., 1 May 2025). The Comptonized component had $\sim 1.41$0 keV and $\sim 1.41$1, indicating a cool, optically thick scattering region (Kashyap et al., 1 May 2025).

A separate IXPE+NuSTAR+Swift/XRT study emphasized the “Western model,” in which soft disk photons are Comptonized in an optically thick, cool corona while a separate thermal component originates from the neutron-star surface or boundary layer (Kumar et al., 6 Apr 2025). Its preferred spectral model,

$\sim 1.41$2

yielded $\sim 1.41$3, compTT seed temperature $\sim 1.41$4 keV, $\sim 1.41$5 keV, optical depth $\sim 1.41$6, and a weak surface Comptonization layer with $\sim 1.41$7 (Kumar et al., 6 Apr 2025). The corresponding Compton $\sim 1.41$8-parameter was reported as $\sim 1.41$9 for compTT and $1.4\,M_\odot$0 for compbb, supporting a picture of nearly saturated scattering in a slab-like configuration and negligible Comptonization in the surface layer (Kumar et al., 6 Apr 2025).

NuSTAR spectroscopy across the Z track demonstrated that simple two-thermal-component fits are inadequate because of a broad asymmetric Fe K$1.4\,M_\odot$1 residual in all states (Coughenour et al., 2017). Comptonized continua improve the fit but do not remove the need for reflection (Coughenour et al., 2017). Full reflection modeling with relconv $1.4\,M_\odot$2 reflionx$1.4\,M_\odot$3 plus diskbb and bbodyrad showed that the inner disk radius remains broadly stable across NB through FB2, with a weighted mean $1.4\,M_\odot$4, corresponding to $1.4\,M_\odot$5 km for a canonical $1.4\,M_\odot$6 neutron star (Coughenour et al., 2017). In that analysis, the Fe line persisted in all states and the source did not show strong evidence for large monotonic disk motion along the track.

The line centroid depends on the model and dataset. NuSTAR diskline fits gave $1.4\,M_\odot$7 keV across the Z track (Coughenour et al., 2017), whereas IXPE+NuSTAR spectro-polarimetry placed the line or diskline near $1.4\,M_\odot$8 keV in some states (Kashyap et al., 1 May 2025), and the XMM-Newton transition study fixed Fe K$1.4\,M_\odot$9 at 6.7 or 6.76 keV depending on the continuum model (R. et al., 5 Feb 2026). A plausible implication is that line centroid differences partly reflect differing continuum prescriptions, ionization assumptions, and energy-band coverage rather than a single directly comparable physical measurement.

4. Timing and cross-correlation phenomenology

Cross-correlation analyses have revealed two distinct lag regimes in GX 349+2. The RXTE archival study used soft and hard bands of $=$0 keV and $=$1 keV, with 16 s bins, and computed discrete CCFs using XRONOS CROSSCOR (Ding et al., 2015). Along the 1998 Z track, 18 of 23 regions showed positive correlations, 5 were ambiguous, and no anti-correlations were found (Ding et al., 2015). Typical cross-correlation coefficients ranged from $=$2 to $=$3, and the corresponding positively correlated lags were short, usually $=$4 s and generally within a few to a few tens of seconds, though one LNB region reached $=$5 s (Ding et al., 2015). Outside that canonical Z-track interval, anti-correlations appeared in 10 observations, positive correlations in 100, and ambiguous behavior in 5 (Ding et al., 2015). The anti-correlated lags spanned tens to thousands of seconds, with examples from $=$6 s to $=$7 s (Ding et al., 2015).

The XMM-Newton study focused instead on $=$8 keV and $=$9 keV light curves binned at 50 s and computed a normalized CCF

$$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$0

To quantify asymmetric peaks, each CCF was modeled with a Gaussian on a linear baseline, described as line + Gaussian + line, and the Gaussian centroid or peak provided the lag estimate (R. et al., 5 Feb 2026). This study found strongly branch-dependent behavior: HB segments showed markedly asymmetric CCFs with relatively low peak correlation coefficients and non-zero hard lags of order a few hundred seconds; NB and FB segments showed highly symmetric CCFs peaking at zero lag; and flux-transition windows exhibited asymmetric CCFs with lags of a few tens to a few hundreds of seconds (R. et al., 5 Feb 2026).

The lag detections in the XMM-Newton work were tested with two independent Monte Carlo approaches. First, 10,000 data-driven pairs of synthetic light curves were generated by perturbing each data point within its measurement uncertainty while preserving variability; second, 10,000 Timmer–König red-noise simulations with $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$1 and the observed mean were used to construct 95% confidence envelopes (R. et al., 5 Feb 2026). For all HB and transition intervals with non-zero lags, the observed CCFs lay outside the 95% confidence bands, and the lag distributions from the data-driven Monte Carlo were centered close to the measured values, leading to the conclusion that the lags are intrinsic at approximately 95% confidence (R. et al., 5 Feb 2026).

These results are not identical to the RXTE phenomenology because the underlying selections differ. RXTE’s on-track analysis covered NB and FB only and emphasized short lags with positive correlations (Ding et al., 2015), while XMM-Newton isolated HB and transition intervals in which longer, asymmetric hard lags became visible (R. et al., 5 Feb 2026). This suggests that the lag behavior of GX 349+2 is highly state- and geometry-dependent, and that datasets lacking HB coverage may systematically miss the longer-lag regime.

5. Interpretations of lags and inner-flow dynamics

The long lags in GX 349+2 have been interpreted in two closely related but not identical ways. In the RXTE study, long-term lags in NS-LMXBs were argued to be inconsistent with short-lag mechanisms such as simple Comptonization or with black-hole-style truncated-disk models, and instead were attributed to an extended accretion-disk corona (ADC) (Ding et al., 2015). In that framework, the corona extends to radii $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$2 km, Comptonizes disk seed photons to produce hard X-rays, and competes with viscous inflow to the neutron-star surface. Two timescales were emphasized: the diffusion or Comptonization escape time,

$$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$3

and the viscous inflow time,

$$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$4

Using $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$5 and $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$6, the paper noted $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$7 s at $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$8 cm and $$(8.7\text{–}19.7~\mathrm{keV})/(6.2\text{–}8.7~\mathrm{keV})$$9 s at $=$0 cm, values compatible with the observed hundreds-to-thousands of seconds lags (Ding et al., 2015).

The XMM-Newton work instead located the relevant dynamics much closer to the star, in the boundary layer or corona near the inner edge of the disk (R. et al., 5 Feb 2026). It argued that hard lags of $=$1 to a few hundred seconds are too long for light-travel or Comptonization delays and are better explained by slow viscous readjustment of the boundary layer: $=$2 Adopting $=$3 km, $=$4 km, $=$5, $=$6, and $=$7 s yields $=$8, described as an extremely low effective viscosity (R. et al., 5 Feb 2026). The paper further proposed a depletion timescale for a low-viscosity boundary layer,

$=$9

and showed that for $2.0\text{–}19.7~\mathrm{keV}$0 the resulting $2.0\text{–}19.7~\mathrm{keV}$1 values reproduce the tens-to-hundreds of seconds lags measured in the HB and during transitions (R. et al., 5 Feb 2026).

The same study estimated the radial extent of the boundary layer from the Popham–Sunyaev scaling

$2.0\text{–}19.7~\mathrm{keV}$2

obtaining $2.0\text{–}19.7~\mathrm{keV}$3 km for one continuum parameterization and $2.0\text{–}19.7~\mathrm{keV}$4 km for another (R. et al., 5 Feb 2026). It also noted that prior XMM-Newton work had suggested a large ADC spanning tens to hundreds of km, and showed that for a thick flow, $2.0\text{–}19.7~\mathrm{keV}$5 gives $2.0\text{–}19.7~\mathrm{keV}$6 km for $2.0\text{–}19.7~\mathrm{keV}$7, $2.0\text{–}19.7~\mathrm{keV}$8 s, and $2.0\text{–}19.7~\mathrm{keV}$9, consistent with an extended corona dynamically coupled to the boundary layer (R. et al., 5 Feb 2026).

Taken together, these interpretations are not mutually exclusive. The RXTE results favor an extended ADC for long-term lags outside the canonical Z track (Ding et al., 2015), whereas the XMM-Newton study attributes branch-dependent HB and transition lags to slow mechanical readjustment or depletion in a low-viscosity boundary layer or inner coronal region (R. et al., 5 Feb 2026). A plausible implication is that different lag classes in GX 349+2 probe different radial zones of the accretion flow.

6. X-ray polarization and accretion geometry

IXPE established that GX 349+2 is weakly but significantly polarized in X-rays. A joint IXPE+NuSTAR study reported a full-exposure $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$0 keV polarization degree $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$1 and polarization angle $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$2, with $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$3 significance (Kashyap et al., 1 May 2025). The same work found energy-resolved values of $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$4 at $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$5 in $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$6 keV, $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$7 at $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$8 in $$(10\text{–}16~\mathrm{keV})/(6.4\text{–}10~\mathrm{keV})$$9 keV, and $\alpha$0 at $\alpha$1 in $\alpha$2 keV, suggesting a possible increase near the Fe-K band (Kashyap et al., 1 May 2025).

A model-independent survey of IXPE Z-source observations gave for GX 349+2 a whole-observation $\alpha$3 keV polarization of $\alpha$4 and $\alpha$5, with $\alpha$6 significance (Gnarini et al., 30 Jun 2025). In that analysis, branch-resolved measurements yielded $\alpha$7, $\alpha$8 in the NB and $\alpha$9, $\sim 0.70$00 in the FB (Gnarini et al., 30 Jun 2025). The PD increase from NB to FB was stated to be consistent within uncertainties at 90% confidence, whereas the PA rotation of approximately $\sim 0.70$01 was reported as significant (Gnarini et al., 30 Jun 2025). This is one of the clearest branch-dependent PA rotations reported among Z sources.

A different IXPE+NuSTAR+Swift/XRT analysis, also with branch-resolved selection, reported significant $\sim 0.70$02 keV polarization in both FB and NB: FB $\sim 0.70$03, $\sim 0.70$04; NB $\sim 0.70$05, $\sim 0.70$06 (Kumar et al., 6 Apr 2025). In the NB it further found an energy-dependent increase from $\sim 0.70$07, $\sim 0.70$08 in $\sim 0.70$09 keV to $\sim 0.70$10, $\sim 0.70$11 in $\sim 0.70$12 keV, implying a PA rotation of approximately $\sim 0.70$13 (Kumar et al., 6 Apr 2025).

These low net polarization degrees are interpreted through Stokes-vector addition of multiple emission components: $\sim 0.70$14 Because disk, boundary/spreading layer, Comptonization, and reflection may have different polarization angles, the net $\sim 0.70$15 can remain near $\sim 0.70$16 even when individual components are polarized at a few percent (Kashyap et al., 1 May 2025). This multi-component dilution is a central result of the IXPE literature on GX 349+2.

The most detailed normal-branch IXPE+NuSTAR spectropolarimetric study concluded that the spectrum is dominated below $\sim 0.70$17 keV by the disk and above $\sim 0.70$18 keV by a compact hotter layer, with energy-dependent polarization strengthening toward high energy (Monaca et al., 9 Jul 2025). In a component-resolved fit, the disk had $\sim 0.70$19 and $\sim 0.70$20, while the compact Comptonizing region had $\sim 0.70$21 and $\sim 0.70$22, with an angular difference of about $\sim 0.70$23 (Monaca et al., 9 Jul 2025). That work stated that the geometry is slightly more favorable to a spreading layer than to a classic boundary layer, although it also emphasized model degeneracies and the difficulty of disentangling reflection polarization (Monaca et al., 9 Jul 2025).

7. Synthesis, limitations, and outstanding issues

Across spectroscopy, timing, and polarimetry, GX 349+2 is consistently described as a near-Eddington neutron-star accretor in which the inner disk reaches close to the star while a substantial boundary or spreading layer and a cool Comptonizing region shape the X-ray output (Coughenour et al., 2017, Kashyap et al., 1 May 2025, R. et al., 5 Feb 2026). The disk does not appear to undergo large monotonic radial excursions through most of the Z track in the NuSTAR reflection analysis, remaining around $\sim 0.70$24 on average (Coughenour et al., 2017). At the same time, XMM-Newton transition spectra requiring a hot disk component with $\sim 0.70$25 keV and relatively low $\sim 0.70$26, together with prior detection of twin kHz QPOs at 712 and 978 Hz, were argued to support an inner flow extending very close to the neutron-star surface or ISCO (R. et al., 5 Feb 2026). These are not strictly identical radius inferences, and the literature treats them with appropriate caution.

A recurrent point of agreement is that branch changes are driven primarily by reorganization of the inner thermal and Comptonizing regions rather than by wholesale movement of the outer observable disk edge (Coughenour et al., 2017, R. et al., 5 Feb 2026). In the timing domain, NB and FB often show tightly coupled soft–hard variability with symmetric or near-zero-lag CCFs, whereas HB segments and transitions reveal asymmetric CCFs with lags of tens to hundreds of seconds (R. et al., 5 Feb 2026). In polarimetry, GX 349+2 remains weakly polarized overall, but shows evidence for energy-dependent and branch-dependent PA rotation, implying that the dominant polarized component changes with spectral state and photon energy (Kumar et al., 6 Apr 2025, Gnarini et al., 30 Jun 2025).

Several limitations remain explicit in the literature. The XMM-Newton lag study is based on a single $\sim 0.70$27 keV observation with 50 s timing bins, and exact lag values and uncertainties are not tabulated despite Monte Carlo validation at approximately 95% confidence (R. et al., 5 Feb 2026). IXPE branch-resolved results depend strongly on state selection, exposure per branch, and whether brief SA or FB excursions are grouped with NB-dominated intervals (Kashyap et al., 1 May 2025, Gnarini et al., 30 Jun 2025, Monaca et al., 9 Jul 2025). Reflection constraints in some flaring states are weakened by high ionization, degeneracies with continuum shape, and spectral mixing across multiple flares (Coughenour et al., 2017). More complex reflection models than simple diskline were attempted in polarimetric studies but often remained unconstrained at current signal-to-noise (Kashyap et al., 1 May 2025).

The present evidence nonetheless supports a coherent picture. GX 349+2 consists of a bright inner disk, a neutron-star boundary or spreading layer with characteristic radii from a few km to tens of km depending on model and definition, an optically thick low-temperature Comptonizing medium, and a persistent Fe-K reflection signature. Its lag behavior implies both fast coupled variability and slower structural readjustments, and its polarization implies weak intrinsic net asymmetry after vector summation of several components. This suggests that GX 349+2 is best understood not through any single diagnostic but through the combined evolution of CCF morphology, continuum decomposition, reflection, and Stokes geometry along the Z track (Ding et al., 2015, Coughenour et al., 2017, Kashyap et al., 1 May 2025, R. et al., 5 Feb 2026).