{"id":"ff5138a1-4608-4b17-b27d-9adfeee986c3","arxiv_id":"2504.20717","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Swift J1727.8-1613 shows a dip at 3 to 15 Hz in the real part of its high-energy cross spectrum, with hard X-ray photons lagging soft photons by between pi/2 and pi.","lead":"An analysis of Insight-HXMT observations of the black hole X-ray binary Swift J1727.8-1613 reveals a dip around 3 to 15 Hz in the real part of the cross spectrum between hard and soft X-rays, where hard photons lag soft ones by more than a quarter of a cycle. The feature is a first for a black hole binary and may help measure the size of the X-ray corona.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Gaussian phase-lag model of Eq. 2 is physically unmotivated and shifts the fitted dip location; the claim that the dip is a distinct 3-15 Hz component, and all derived evolution, energy dependence, and size results, are conditional on this lag law.","rationale":"The reader's verdict is CONDITIONAL with high confidence, and I agree with the identified weakest assumption. The data-level dip is a reasonable detection: it is visible in the averaged, unrotated real part of the cross spectrum (Fig. 5) before any lag model is applied, and it persists across the grouped observations used for Fig. 10. However, the paper's stronger interpretive claims, namely a distinct new Lorentzian component, its evolution and energy dependence, and a greater-than-10,000 km corona, are not model-independent. The same data fitted with the constant phase-lag model require a dip Lorentzian with a centroid below 0.81 Hz rather than 3.6 Hz (Table 3), demonstrating that the feature's parameters are degenerate with the assumed lag law. The authors explicitly acknowledge the lack of justification for the Gaussian profile in Section 5.1. This does not invalidate the detection, but it means the component-level results should be presented as conditional on a phenomenological lag model. The proposed concrete test either hardens the component identification or shows it to be an artifact of the adopted lag law. Therefore the reader's conditional-accept verdict stands unchanged.","tokens_in":23859,"tokens_out":10608,"duration_ms":117763,"concrete_test":"Re-fit the Group #9 PDS and cross-spectrum data with the Gaussian phase-lag model but leave sigma_dip free instead of tying it to the dip Lorentzian FWHM, and also with an alternative smooth lag profile (e.g., a Lorentzian phase-lag function or a low-order polynomial in frequency for all components). If the dip Lorentzian remains necessary with a centroid near 3-15 Hz and DeltaPhi_D near 2.7 rad under both choices, the component is robust; if the negative real part is absorbed by the other components' lag profiles or the dip moves or disappears, the Gaussian-based component and the size estimate are model-dependent.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The direct detection of negative Re(CS) in the averaged Group #9 data (Fig. 5) is not in doubt. What is at risk is the identification of that feature with a separate Lorentzian component at about 3-15 Hz. That identification is made through the Gaussian phase-lag model introduced in Section 4.1.2 (Eq. 2, with sigma_i tied to the Lorentzian FWHM) specifically to escape the degeneracy of the constant phase-lag model (Section 4.1.1). Table 3 shows the dip Lorentzian centroid moves from <0.81 Hz under the constant phase-lag model to 3.6 +/- 0.5 Hz under the Gaussian phase-lag model, and Table 2 shows similar shifts for the representative observation. The authors state in Section 5.1 that there is little justification for a Gaussian frequency dependence of phase lags. Because the dip's evolution (Fig. 10), energy dependence (Fig. 13), and the coronal size estimate from tau = DeltaPhi_D/(2*pi*nu0) in Section 5.3 are all read off this Lorentzian, a different lag law could change the component's parameters or absorb the dip into other components. The central detection survives, but the 'new component' and the size claim do not follow model-independently.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents Insight-HXMT timing observations of the black-hole candidate Swift J1727.8-1613 during its 2023 outburst, focusing on the Normal State defined by Yu et al. (2024). The authors report a dip-like feature at ~3-15 Hz in the real part of the high-energy (28-200 keV) vs low-energy (2-10 keV) cross spectrum, which reaches negative values near its minimum and corresponds to a phase lag between pi/2 and pi. They jointly fit the two power spectra and the real and imaginary parts of the cross spectrum with a multi-Lorentzian model under three lag prescriptions: constant phase lag, constant time lag, and a Gaussian phase-lag model (Eq. 2). For the co-added Group #9 data, the Gaussian model gives chi^2/dof = 781/678 and places the dip Lorentzian at 3.6 +/- 0.5 Hz, whereas the constant phase-lag model gives a centroid < 0.81 Hz. Using the Gaussian model results, the authors track the dip's evolution with QPO frequency, its energy dependence, derive a coronal size estimate that consistently exceeds 10,000 km, and compare the energy-dependent rms and phase-lag shapes with predictions of the time-dependent Comptonization model vKompth.","tokens_in":24206,"tokens_out":7987,"duration_ms":78402,"significance":"If the central observational claim holds, this is a genuinely new finding: a strong, high-energy dip in the real part of the cross spectrum with a phase lag exceeding pi/2 in a black-hole X-ray binary has not been reported before. The direct detection is based on a large number of averaged segments (547 for Group #9), and the joint fitting of PDS and cross-spectrum components follows a published, reproducible method (Mendez et al. 2024). The predicted phase-lag and coherence spectra derived from the joint fit are compared with the data rather than fitted, which is a strength. The main weakness is that the quantitative interpretation of the dip as a distinct ~3-15 Hz variability component, and all derived quantities based on it, depend on an ad hoc Gaussian phase-lag model whose physical justification the authors themselves question in Section 5.1. The direct detection of negative Re(CS) is model-independent; the component identification and the size estimate are not.","major_comments":[{"comment":"The identification of a distinct dip Lorentzian at 3-15 Hz is strongly model-dependent. Under the constant phase-lag model the dip centroid is < 0.81 Hz, while under the Gaussian phase-lag model it becomes 3.6 +/- 0.5 Hz (Table 3). The Gaussian prescription in Eq. (2) is introduced to break a degeneracy but, as the authors state in Section 5.1, there is little physical justification for a Gaussian frequency dependence of phase lags. Because the dip's evolution (Section 4.3, Fig. 10), its energy dependence (Section 4.4, Fig. 13), and the coronal size estimate from tau = DeltaPhi_D/(2*pi*nu0) in Section 5.3 are all read off the parameters of this Lorentzian, those quantitative results are not robust to the assumed lag law. I recommend testing at least one additional, differently shaped lag prescription (e.g., a Lorentzian or power-law frequency dependence) and explicitly reporting which conclusions survive across lag models, or reframing the paper so that the model-dependent decomposition is clearly separated from the model-independent direct detection.","section":"§4.1.2, Eq. (2); Table 3; §5.1"},{"comment":"The central claim that the real part of the cross spectrum reaches negative values near the dip minimum is not quantified statistically. The paper does not report the value and uncertainty of Re(CS) at the dip minimum, nor a confidence interval for the phase lag exceeding pi/2, computed directly from the averaged segments. Adding such an estimate (e.g., the significance of the most negative frequency bin, or a simple bootstrap error on the phase lag in the 3-15 Hz band) would let the reader judge the detection without relying on the model fits. This is important because the title and abstract present the negative excursion as the paper's headline result.","section":"§3.2; Fig. 5"}],"minor_comments":[{"comment":"The MCMC description says 'a total of 200000 samples and a burn-in phase of 200000', which leaves zero post-burn-in samples. Please clarify whether the chain length is 400000 in total with 200000 burn-in, or whether the quoted numbers are already the post-burn-in length.","section":"§4.4"},{"comment":"The y-axis label 'Maximum phase lag (rad)' appears inconsistent with the text: the text quotes values of 0.4*pi to 0.9*pi, while the axis ticks run from 0.375 to 1. If the ticks are in units of pi, the label should read e.g. 'Maximum phase lag (pi rad)'; if they are in radians, the tick values and text do not agree.","section":"Fig. 10"},{"comment":"The Gaussian phase-lag model is described as giving a 'slightly better reduced chi^2' than the constant phase-lag model, but for Group #9 the difference is actually Delta_chi^2 = 29 for the same number of parameters (781 vs 810 for 678 dof). Consider stating this as a formal model comparison rather than 'slightly better'.","section":"Abstract and §4.2"},{"comment":"The comparison with vKompth predictions is qualitative ('closely resemble'). Since the paper does not fit vKompth to the data, I suggest making explicit that this is a qualitative similarity and, if possible, stating which observable would quantitatively distinguish the low-feedback scenario.","section":"§5.3"}],"recommendation":"major_revision","confidential_remarks":"The direct detection of the dip-like feature in the real part of the cross spectrum is the paper's core observational contribution and appears solid, but the manuscript's more ambitious claims - a new variability component at 3-15 Hz and a coronal size estimate - rest on an ad hoc Gaussian phase-lag model whose parameters shift the component centroid by a factor of several relative to the constant phase-lag model. I would like to see either a robustness analysis with alternative lag laws or a substantial softening of the model-dependent claims before this paper can be accepted. The extensive discussion of jet sizes and vKompth, while interesting, should be clearly separated from the robust observational results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The dip detection is real and new. Averaging 547 segments in Group #9 makes the negative real part of the cross spectrum at 3-15 Hz hard to dismiss; no prior black hole binary has shown phase lags that cross pi/2 in a broad hump. The analysis is careful: joint fits to PDS and real/imaginary CS, three lag models, residuals shown, and the energy dependence was checked with two different lag models, giving similar trends. The authors are also honest in Section 5.1 that there is little physical justification for the Gaussian phase-lag profile.\n\nThe soft spot is exactly the one flagged by the stress test. Under the constant phase-lag model the dip Lorentzian centroid is <0.81 Hz, under the Gaussian model it is 3.6 +/- 0.5 Hz. That discrepancy is large enough to undercut the claim that the dip is a distinct 3-15 Hz component. The Gaussian model is preferred by only Delta chi^2 ~ 29 with the same degrees of freedom, which is not compelling, and it was introduced primarily to escape degeneracy. Everything that follows - the evolution of the dip frequency, the energy-dependent phase lag, and the size estimate - is read off this model-dependent Lorentzian. The size estimate L ~ tau c with tau = DeltaPhi_D/(2 pi nu0) is a one-line simplification, and the abstract states \"consistently exceeds 10,000 km\" without noting that it assumes the Gaussian lag law and a light-crossing interpretation. The vKompth comparison is also qualitative, a resemblance rather than a fit.\n\nNone of this invalidates the central detection. The negative real part is present in the averaged data regardless of model. What needs fixing is the framing: the 3-15 Hz dip as a clean new component and the coronal size are conditional results, not direct measurements. A revised version that presents the dip parameters under both lag models side by side and moves the size estimate to a clearly labeled, model-dependent section would be much stronger.\n\nI would send this to peer review. It is a solid observational paper with one load-bearing but acknowledged caveat, and the community needs the detection on record. I would also cite it, with a caveat about the model dependence. Worth a reading group discussion on lag models.","headline":"The negative-going dip in the cross spectrum is a genuine, new detection for a black hole X-ray binary, but its interpretation as a distinct 3-15 Hz component and the coronal size estimate depend heavily on a lag model the authors themselves admit is physically unmotivated.","tokens_in":24773,"tokens_out":2021,"would_cite":true,"duration_ms":23304,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper reports the first detection of a dip at 3–15 Hz in the real part of the cross spectrum of the black-hole binary Swift J1727.8–1613, with hard photons lagging soft ones by more than $\\pi/2$.","keywords":["black hole X-ray binaries","cross spectrum","phase lags","quasi-periodic oscillations","accretion flow","corona","Comptonization","Insight-HXMT"],"falsifier":"Re-fit the same Insight-HXMT power and cross spectra with the phase-lag frequency dependence left free (e.g., a flexible spline or a sum of Gaussians) and check whether a 3–15 Hz Lorentzian with phase lag between $\\pi/2$ and $\\pi$ remains significant; alternatively, measure the dip's energy-dependent phase-lag and rms spectra with higher photon statistics and test whether the pivot stays near 15 keV with the rms minimum coinciding with the phase-lag drop.","tokens_in":23663,"feed_emoji":"🔭","tokens_out":7887,"duration_ms":69444,"temperature":0.7,"pith_summary":"The paper reports the first detection, in a black-hole X-ray binary, of a dip at about 3–15 Hz in the real part of the cross spectrum between hard (>25 keV) and soft (<10 keV) photons, seen by Insight-HXMT in Swift J1727.8–1613 during its 2023 outburst. At the dip's minimum the real part becomes negative, meaning the hard photons lag the soft ones by more than a quarter cycle but less than half a cycle. The authors fit the power spectra and the real and imaginary parts of the cross spectrum simultaneously with a multi-Lorentzian model and show that a Gaussian frequency-dependent phase-lag law describes the dip best. From that Lorentzian they estimate the size of the emitting region, which stays above 10,000 km as the QPO frequency rises from 0.13 to 2.0 Hz, and they find a break near 15 keV in both the phase-lag and fractional-rms energy spectra. The energy dependence resembles the vKompth time-dependent Comptonization model with a low feedback factor, pointing to the corona as the origin of the new component.","feed_headline":"First dip in a black-hole binary's cross spectrum","feed_subtitle":"Swift J1727.8: hard X-rays lag soft ones by >90 degrees in a 3-15 Hz band, implying a corona above 10,000 km.","key_machinery":"The carrying machinery is a joint fit of the power density spectrum and the real and imaginary parts of the cross spectrum, under the assumption that each Lorentzian component is incoherent with the others but fully coherent between energy bands, together with a new lag model: the phase lag of each Lorentzian is a Gaussian function of Fourier frequency, $g_i(\\nu)=2\\pi k_i \\exp[-\\frac12(\\nu-\\nu_{0,i})^2/\\sigma_i^2]$, with $\\sigma_i$ tied to the Lorentzian FWHM. Unlike the constant phase-lag model, this keeps the dip Lorentzian's negative contribution confined to the 3–15 Hz range and removes the need for compensating Lorentzians, which is why the fitted dip parameters (centroid $3.6\\pm0.5$ Hz, FWHM $10.1\\pm0.3$ Hz, and $\\Delta\\Phi_D=2.78^{+0.08}_{-0.07}$ rad for Group 9) carry the argument. The same fitted Lorentzian is used to estimate the corona size through $L\\sim c\\tau$ with $\\tau=\\Delta\\phi_D/(2\\pi\\nu_0)$, giving sizes consistently above 10,000 km.","core_discovery":"On its own terms, the paper establishes that Swift J1727.8–1613 shows a variability component never seen before in a black-hole binary: a dip in the real part of the cross spectrum between 28–200 keV and 2–10 keV photons, spanning roughly 3–15 Hz, with the real part crossing zero and reaching negative values near the minimum. Because the imaginary part stays positive there, the corresponding phase lag lies between $\\pi/2$ and $\\pi$. The dip is caused by a change in phase lag rather than in the modulus of the cross spectrum. Jointly fitting the power spectra and the cross spectrum with 13 Lorentzians locates the feature in a broad Lorentzian whose centroid frequency rises from below 1 Hz to about 6 Hz as the QPO frequency increases, and its phase lag at the centroid grows from below $0.4\\pi$ to about $0.9\\pi$ before flattening. The same Lorentzian shows a break near 15 keV in its energy-dependent phase lag and fractional rms, matching the qualitative shape predicted by the vKompth low-feedback Comptonization model.","pith_inferences":["My inference: the dip may be a generic feature of the hard-intermediate state in black-hole binaries that was missed because in previously studied sources the phase-lag hump stays below $\\pi/2$; sources like MAXI J1820+070 and XTE J1550−564 should be re-examined for a negative real part.","My inference: if the Gaussian lag law is a stand-in for a physical lag that peaks at the Lorentzian centroid, the dip centroid frequency may track the QPO frequency in a universal way, giving a testable scaling from the paper's evolution plot.","My inference: the vKompth resemblance suggests the seed photons for the dip component differ from typical disc seeds; spectral-timing fits that leave the seed temperature free could confirm whether a distinct Comptonizing region is responsible."],"forward_implications":["If the dip is real, the cross spectrum separates a new variability component in the 3–15 Hz band with hard lags exceeding $\\pi/2$, a property not seen in other black-hole binaries, and it constrains Comptonization geometry.","The size estimate above 10,000 km (up to roughly $10^4\\,R_g$ for an 8-solar-mass black hole) implies the Comptonizing region is extended rather than compact.","The break near 15 keV in both phase lag and fractional rms provides an energy scale that the vKompth low-feedback model can reproduce, linking timing and spectral modeling of the corona.","The coherence function's decline at high energies indicates that different variability components dominate the soft and hard bands, supporting the presence of multiple radiative regions."],"supporting_citations":[{"why":"Supplies the joint-fitting method for power spectra and the real and imaginary parts of the cross spectrum, along with the constant phase-lag and constant time-lag models.","marker":"Méndez et al. (2024)"},{"why":"Provides the same Insight-HXMT observations and the Normal State / Flare State classification that defines the data selection here.","marker":"Yu et al. (2024)"},{"why":"Introduces the Gaussian phase-lag parameterization adopted as the preferred lag model.","marker":"Zhou et al. (2022)"},{"why":"Provides the vKompth time-dependent Comptonization model whose predicted rms and lag spectra are compared with the dip's energy dependence.","marker":"Karpouzas et al. (2020)"},{"why":"Presents the vKompth calculations for low feedback factors that the paper's energy-dependent spectra resemble.","marker":"Bellavita et al. (2022)"},{"why":"Gives the light-travel-time relation $L\\sim c\\tau$ used to convert the dip's phase lag into a corona size.","marker":"Reynolds & Nowak (2003)"},{"why":"Documents a phase-lag hump below $\\pi/2$ in MAXI J1820+070, the closest prior feature and the main comparison case.","marker":"Kawamura et al. (2023)"},{"why":"Documents a similar hump in XTE J1550−564 with phase lag always below $\\pi/2$, providing the contrast for the new negative real part.","marker":"Cui et al. (2000)"},{"why":"Defines the cross-spectrum and coherence formalism used throughout the analysis.","marker":"Nowak & Vaughan (1996)"}],"fun_headline_variants":["Black-hole binary shows first cross-spectrum dip","Swift J1727: hard X-rays lag soft by >90°","New dip in Swift J1727 cross spectrum hints corona size","First phase-lag dip seen in black-hole binary","Swift J1727's 3-15 Hz dip: phase lag reveals corona"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quantitative dip parameters and the derived corona size rest on the assumption that each Lorentzian's phase lag follows a Gaussian curve peaked at its centroid frequency; the paper itself notes there is little physical justification for that specific frequency dependence, and a different lag law would change the amplitude, width, and inferred size.","fun_headline_variants_meta":{"raw":{"variants":["Black-hole binary shows first cross-spectrum dip","Swift J1727: hard X-rays lag soft by >90°","New dip in Swift J1727 cross spectrum hints corona size","First phase-lag dip seen in black-hole binary","Swift J1727's 3-15 Hz dip: phase lag reveals corona"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00068,"raw_usage":{"total_tokens":3197,"prompt_tokens":1157,"completion_tokens":2040,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":773,"completion_tokens_details":{"reasoning_tokens":1953}},"tokens_in":773,"tokens_out":2040,"duration_ms":14899,"temperature":1.0,"reasoning_tokens":1953,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T05:22:35.267575+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-fit the same Insight-HXMT power and cross spectra with the phase-lag frequency dependence left free (e.g., a flexible spline or a sum of Gaussians) and check whether a 3–15 Hz Lorentzian with phase lag between $\\pi/2$ and $\\pi$ remains significant; alternatively, measure the dip's energy-dependent phase-lag and rms spectra with higher photon statistics and test whether the pivot stays near 15 keV with the rms minimum coinciding with the phase-lag drop.","supporting_citations":[],"review_version":1}