Cross-spectral Analysis of the Type-C Quasi-periodic Oscillation Shoulder Component in GX 339-4

Abstract: We revisit Rossi X-ray Timing Explorer (RXTE) observations of GX~339$-$4 during the rising phase of its 2006/2007 outburst and apply a joint power-density-spectrum (PDS)--cross-spectrum (CS) decomposition to the type-C quasi-periodic oscillation (QPO) region. Within this framework, the QPO region is described by a narrow QPO fundamental and a neighboring high-frequency shoulder, whose amplitudes and phase lags can be measured separately. The shoulder is first detected at MJD~54142.04, mainly through the imaginary part of the CS and a narrow local structure in the phase-lag spectrum, before becoming a resolved high-frequency shoulder in the PDS. It follows the QPO frequency evolution on the high-frequency side, with $R_ν=ν<em>{\rm sh}/ν</em>{\rm QPO}\simeq1.04$--$1.18$. The QPO lag remains small, typically below $\sim0.17$~rad, whereas the shoulder carries a larger hard lag of $\sim0.5$--$0.8$~rad. Energy-resolved fits show the same separation: the QPO lag is close to zero or only weakly positive across most of the energy band, while the shoulder lag is systematically larger and generally increases with photon energy. The two components have broadly similar rms--energy shapes, although their relative strengths evolve during the observed sequence. Although the shoulder remains broad, with $Q\sim2$--$4$, its lag and rms--energy behavior resemble those of the type-B QPO detected shortly after our observations. This similarity raises the interesting possibility that the shoulder is related to an earlier, broader stage of the variability later seen as the type-B QPO.

• The paper identifies a distinct high-frequency shoulder beside GX 339−4’s type-C QPO using joint power-density and cross-spectral decomposition, revealing phase lags of about 0.5–0.8 radians versus less than 0.17 radians for the fundamental.
• The shoulder emerges near 2.90 Hz, remains close to the QPO frequency with a ratio of 1.04–1.18, and cannot be explained by intra-observation frequency drift because the measured centroid scatter is roughly seven times smaller than the component separation.
• The findings show that interval-averaged QPO lags can be substantially biased by blended components and suggest that the shoulder may be a broader hard-intermediate-state precursor to the later Type-B QPO.

Overview and motivation

This paper revisits the 23 Rossi X-ray Timing Explorer (RXTE) observations of the black-hole X-ray binary GX 339−4 taken during the rising phase of its 2006/2007 outburst, previously analyzed by Zhang et al. (2017) for their type-C quasi-periodic oscillation (QPO) phase lags. The authors apply the joint power-density-spectrum (PDS)–cross-spectrum (CS) multi-Lorentzian decomposition introduced by Méndez et al. (2024, M24), in which each Lorentzian component has tied centroid frequency and width across the PDS and both CS projections, but independent cross-spectral normalization and phase lag. The central result is that the type-C QPO region is not a single feature: it consists of a narrow QPO fundamental plus a neighboring high-frequency shoulder that carries a phase lag several times larger than that of the fundamental.

The motivation is methodological as much as astrophysical. Conventional QPO lags are obtained by averaging the real and imaginary parts of the CS over an interval around the QPO centroid, typically one FWHM wide. When two nearby components with different phases contribute to that interval, the measured lag is the phase of the summed cross-vector rather than the intrinsic lag of either component. The paper quantifies how strongly this biases lag measurements in GX 339−4 and traces the emergence of the shoulder component across the outburst rise.

Data and method

The analysis uses RXTE/PCA data with 128-s segments at $1/1024$ s time resolution (Nyquist frequency 512 Hz), geometrically rebinned by a factor $\approx 10^{1/100}$, in fractional rms normalization with Poisson noise subtracted. The main two-band analysis uses soft (2–5.7 keV) and hard (5.7–115 keV) bands; energy-resolved fits use six PCA bands with 2.0–5.7 keV as reference, restricted to 0.01–20 Hz. Fits were performed in XSPEC after rotating the CS counterclockwise by 45°, which equalizes the amplitudes of the two fitted projections for small lags without altering recovered parameters. Each Lorentzian adopts a constant phase-lag prescription, $\Delta\phi_i = 2\pi k_i$, adequate for narrow features over limited frequency ranges where no phase wrapping occurs.

A key classification distinction is drawn between components resolved in the full-band PDS and "CS-selected" components—those weak or unresolved in the PDS but required by structured residuals in the imaginary part of the CS or the derived phase-lag spectrum. The phase-lag spectrum and intrinsic coherence are not fitted; they are predicted from the best-fitting PDS–CS model, providing an internal consistency check.

Emergence and evolution of the shoulder

The shoulder first appears at MJD 54142.04 (Obs 15). An eight-Lorentzian model already gives an acceptable fit to the full-band PDS with no residual structure near the QPO peak ($\nu_{\rm QPO}\simeq2.45$ Hz). However, this model leaves a structured residual in the imaginary part of the CS and fails to reproduce a narrow local enhancement in the phase-lag spectrum. Adding a ninth Lorentzian improves the joint fit from $\chi^2=818.8$ (752 dof) to $\chi^2=780.0$ (746 dof), with the improvement concentrated precisely in the imaginary CS and local lag structure. A tenth Lorentzian adds nothing significant. This observation is therefore classified as transitional: the shoulder is CS-selected but not independently resolved in the PDS.

By Obs 17 (MJD 54143.87), the shoulder is resolved in the full-band PDS as well, and it remains detectable through the end of the sequence. Across all detections:

Property	QPO fundamental	Shoulder
Frequency range	0.14 → ~5–6 Hz	2.90 → ~6 Hz
Frequency ratio $R_\nu=\nu_{\rm sh}/\nu_{\rm QPO}$	—	1.04–1.18
Phase lag	≲0.17 rad	~0.5–0.8 rad
Fractional rms	~4–12%	~3–6%
Quality factor	high (narrow)	$Q\sim2$–4

The lag separation $\Delta\phi=\phi_{\rm sh}-\phi_{\rm QPO}\simeq0.4$–0.7 rad exceeds $4\sigma$ in every detection. The rms²-weighted vector-combined lag over the QPO region lies at ~0.2–0.35 rad, intermediate between the two components—directly illustrating the bias inherent in interval-averaged lag measurements once the shoulder is present.

The near-unity frequency ratio rules out identification with the harmonic sequence and indicates that both components are coupled to the same evolving accretion-flow time-scale, consistent with the known correlated motion of Lorentzian components through state transitions.

Excluding frequency drift

A natural concern is whether the shoulder is an artifact of intra-observation QPO centroid drift smearing the profile. The authors address this directly with a dynamical PDS of Obs 16: fitting 24 individual 128-s segments yields a mean centroid of $\approx 10^{1/100}$0 Hz with intrinsic rms scatter $\approx 10^{1/100}$1 Hz, i.e., only ~2.5% of the mean frequency—roughly seven times smaller than the ~0.61 Hz QPO–shoulder separation. Complementarily, short-interval (256-s) cross spectra show the shoulder persisting in the imaginary part at the same frequency. These tests argue against temporal frequency drift as the sole origin, although the authors concede that sub-threshold variations below the sensitivity of individual PDS remain unresolved, and that energy-dependent centroid or nonstationary transfer-function effects are not excluded. They note that wavelet or Hilbert–Huang analyses would be needed for a definitive characterization, deferring this to future work.

The coherence spectrum provides independent support: when both components are present, coherence peaks near the QPO and shoulder frequencies with a shallow dip between them—the signature expected when overlapping components with different phases partially cancel their cross-vectors. This mirrors the behavior reported for simultaneous Type-B/Type-C QPOs in Swift J1727.8−1613. In observations without a shoulder, coherence stays near unity across the QPO region.

Energy dependence

Energy-resolved fits reinforce the component separation. Relative to the 2.0–5.7 keV reference band, the QPO lag remains small (≲0.2 rad) across most of the PCA band, consistent with zero or slightly negative at the highest energies in several observations, while retaining structure around ~6–8 keV that plausibly reflects a reflected-emission contribution. The shoulder instead shows a systematically larger hard lag that generally increases with photon energy, from ~0.3–0.5 rad in the lowest non-reference band to ~0.8–1.7 rad above 20 keV. The authors caution that the highest-energy points have the largest uncertainties due to declining PCA count rate and should not be over-interpreted individually—though the QPO–shoulder separation holds at all energies where the shoulder is resolved.

The rms–energy spectra of the two components are broadly similar, rising toward ~10–20 keV before flattening, consistent with variability dominated by Comptonized emission diluted at low energies by the thermal disc. Their relative strengths evolve: the shoulder rms starts well below the QPO's, becomes comparable by Obs 20 (~9–13% around 8–20 keV), briefly exceeds it in Obs 21, then weakens again.

Implications for prior lag measurements and the Type-B connection

Two consequences follow directly. First, since the strongest hard lag in the QPO region belongs to the shoulder rather than the fundamental, lag-energy spectra measured over a fixed QPO interval—such as those of Zhang et al. (2017)—blend two components with different phases, weighted by both frequency and energy. The comparison between the two studies is necessarily qualitative, given also the different reference band (2–4 keV versus 2.0–5.7 keV).

Second, the shoulder appears near the end of the hard-intermediate state, shortly before a Type-B QPO at ~6.7 Hz was reported in the following observation. The shoulder's lag–energy behavior (hard lags increasing with energy) broadly matches that of this Type-B QPO in nearly all detections, and its rms–energy shape progressively resembles the Type-B pattern. The resemblance extends qualitatively to the dual-corona Comptonization framework proposed for Type-B QPOs. The authors are careful about classification: with $\approx 10^{1/100}$2–4, appearing on the high-frequency side of the Type-C QPO rather than as an isolated narrow peak, the shoulder does not meet canonical Type-B criteria on PDS morphology alone. The claim is correspondingly modest—that the shoulder may represent an earlier, broader stage of the variability later seen as the Type-B QPO—an interpretation supported by analogous HIMS-shoulder-to-SIMS-Type-B evolution reported in Swift J1727.8−1613, and by observations of simultaneously coexisting Type-B and Type-C QPOs during transitions.

Limitations and open questions

Several caveats bound the results. The constant phase-lag prescription is a simplification, justified only for narrow features over limited frequency ranges. The physical origin of the shoulder remains undetermined: delayed feedback in an extended Comptonizing region and inward propagation of accretion-rate fluctuations are both compatible with the data, and RXTE cannot distinguish them. The highest-energy lag measurements carry large uncertainties. Sub-resolution frequency variability could still affect component widths. Finally, the Type-B precursor interpretation rests on qualitative spectral-timing similarity rather than a demonstrated evolutionary mechanism, leaving open whether the shoulder continuously narrows into the Type-B QPO or the two are merely related phenomena.

Conclusion

Applying joint PDS–CS decomposition to the 2006/2007 rising outburst of GX 339−4, this work establishes that the type-C QPO region contains a distinct high-frequency shoulder component, first revealed through the imaginary part of the cross spectrum at MJD 54142.04 and later resolved in the PDS. The shoulder tracks the QPO frequency at $\approx 10^{1/100}$3–1.18 yet carries a hard lag of ~0.5–0.8 rad against the QPO's ≲0.17 rad—a >4σ separation in every detection—and dominates the hard lag of the QPO region at high photon energies. The results demonstrate concretely that conventional interval-averaged QPO lags can be biased by neighboring blended components, and they raise a specific, testable hypothesis: that the Type-B QPO emerges gradually from a broader HIMS-stage precursor rather than appearing abruptly at the SIMS transition.