{"id":"1821a0d9-889b-47c1-870c-7c943b5a8eff","arxiv_id":"1908.06099","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A Gaussian-process-based method for measuring time lags in gapped X-ray light curves yields the first low-frequency iron K reverberation lag in the radio-loud AGN 3C 120, implying a coronal height of about 13 gravitational radii.","lead":"This paper builds a Gaussian process framework that fills in gaps in X-ray light curves so timing analysis can be done at lower frequencies than usual, and applies it to the radio-loud galaxy 3C 120. It reports the first low-frequency X-ray reverberation detection in this source, with the iron K line lagging the continuum by about 3800 seconds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reverberation claim rests on rejecting a hard-lag continuum by shape alone; no hard-lag model is fit to the lag-energy spectrum, so the 3800 s lag may not measure light travel time.","rationale":"The reader's weakest assumption is well targeted: the central claim that the lag-energy spectrum reveals reverberation, and hence a coronal height, is only as strong as the exclusion of a continuum hard lag. The paper's exclusion is purely qualitative; no competing hard-lag model is fit, and the data band is chosen after seeing the low-frequency lag in broad bands. The 7–10 keV drop is the strongest piece of evidence against a simple monotonic hard lag, but because the lag-energy uncertainties are large and the GP method is known to introduce frequency-dependent systematic biases in the very band used, the shape alone is not decisive. I agree with the reader's conditional verdict: the method and measurement are plausible and worth publishing, but the physical interpretation should be tested against a quantitative hard-lag model before being accepted. No additional concern seemed more load-bearing than this one, and the paper's extensive simulations, including null tests, provide real independent support for the GP machinery itself. The appropriate action is therefore to keep the conditional recommendation rather than move to accept or reject.","tokens_in":24462,"tokens_out":8920,"duration_ms":96425,"concrete_test":"Fit a two-parameter hard-lag model τ(E) = a ln(E/2 keV) + b (or a power law) to the lag-energy measurements in Fig. 11, accounting for the quoted asymmetric errors, and compare via Δχ² against a simple reverberation model (e.g., a relativistic line response plus a constant continuum lag). Additionally, run the GP pipeline on simulated 3C120-like light curves containing a pure hard-lag continuum with no reverberation (τ(E) = a ln E) and with the same count rates, gaps, and segment lengths, then compute the recovered lag-energy spectrum. If the hard-lag model either fits the measured spectrum or produces an iron-K-like peak with a 7–10 keV drop in simulation, the reverberation interpretation and the implied coronal height are not uniquely supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central physical claim — that the 5–7 keV band lags 1.6–2 keV by (3800 ± 1500) s because of disc reverberation, implying a corona at (13 ± 8) rg — depends on excluding a continuum hard lag. Section 5.3 dismisses the hard-lag alternative with one qualitative sentence: \"A hard lag within the continuum emission itself would have produced a lag-energy spectrum smoothly increasing with energy which is not seen here.\" No quantitative hard-lag model is fitted to the measured lag-energy points in Fig. 11. The observed drop in the 7–10 keV bin does argue against a simple monotonic hard lag, but the error bars are large and the GP systematic errors in the 5×10^-6–3×10^-5 Hz band are frequency- and energy-dependent (Section 4.1.2), so the shape could be distorted. A hard-lag model of the form τ(E) = a ln(E/E0) + b could plausibly fit the 1.6–2, 3–4, 4–5, and 5–7 keV points within their 1σ uncertainties; the 7–10 keV point would need explanation via band contamination, but that is not tested. Without fitting this alternative, the 3800 s lag is not uniquely attributable to light travel time, and the coronal-height estimate is not secured. Post-hoc selection of the frequency band, and the model-internal 99.5% 'significance' based on GP posterior samples, further inflate confidence.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces a Gaussian-process framework for Fourier-domain X-ray timing on light curves with orbital gaps. The method fits a kernel (rational quadratic or Matern-1/2) to each energy-band light curve, draws continuous sample light curves, and computes lag-frequency and lag-energy spectra from the samples. Simulations with injected lags validate that phase relationships and lag amplitudes are recovered with known systematic and statistical errors, including for low-Earth-orbit geometries and for long XMM-Newton orbits. The framework is applied to three 2013 XMM-Newton segments of the radio-loud AGN 3C 120. The paper reports a lag-frequency spectrum in which the 4-7 keV band lags the 1.2-4 keV band at low frequencies, and a lag-energy spectrum over 5x10^-6 to 3x10^-5 Hz whose shape is interpreted as reverberation from the inner disc. The 5-7 keV iron-K band is found to lag the 1.6-2 keV continuum by (3800 +/- 1500) s, implying a coronal height of (13 +/- 8) rg. The claimed physical result is a first reverberation measurement in a radio-loud AGN.","tokens_in":24746,"tokens_out":5171,"duration_ms":46557,"significance":"The Gaussian-process timing method is a potentially valuable contribution: it is implemented in a public package (pylag), validated on simulations with known lags, includes a null test for zero lag, and marginalizes over hyperparameter uncertainties with MCMC. If the 3C 120 result is correct, it extends reverberation measurements to higher-mass radio-loud AGN and provides a direct probe of the corona-jet connection. However, the astrophysical claim currently rests on a qualitative exclusion of the hard-lag continuum interpretation, a post-hoc frequency-band choice, and a significance measure that is internal to the GP posterior rather than calibrated against a null hypothesis. These issues are fixable but are load-bearing for the central conclusion.","major_comments":[{"comment":"The dismissal of the continuum hard-lag alternative is qualitative rather than quantitative. The sentence 'A hard lag within the continuum emission itself would have produced a lag-energy spectrum smoothly increasing with energy which is not seen here' is not supported by a fit of any hard-lag model to the measured lag-energy points. A model such as tau(E) = a ln(E/E0) + b should be fitted to the points in Fig. 11, with the 7-10 keV bin treated via band contamination, and the goodness of fit compared with the reverberation interpretation. Without this test, the (3800 +/- 1500) s lag is not uniquely attributable to light travel time, and the derived coronal height of (13 +/- 8) rg is not secured.","section":"Section 5.3, Fig. 11"},{"comment":"The frequency band (5x10^-6 to 3x10^-5 Hz) over which the lag-energy spectrum is computed appears to be selected after inspecting the same data: the caption of Fig. 11 describes this range as 'where the iron K band is seen to lag behind the energy band dominated by the X-ray continuum'. This post-hoc selection is not corrected for in the reported significance. The paper should apply the identical selection procedure to null and alternative simulations and report the resulting false-alarm rate, or otherwise account for the trials factor.","section":"Sections 5.2-5.3, Figs 9 and 11"},{"comment":"The 99.5 per cent significance is defined as the percentage of GP posterior samples in which the 5-7 keV bin responds later than the 1.6-2 keV bin. Because the GP hyperparameters are fit to the same observed light curves and the samples are conditional on the observed data, this is not a frequentist detection probability. The significance should be calibrated with simulations of lag-free light curves processed through the same GP pipeline, and the false-alarm probability should be quoted.","section":"Section 5.3, significance paragraph"},{"comment":"The validation simulations show systematic errors of order 10-20 per cent in the lowest-frequency bins for the 3C 120-like observing geometry (Fig. 7), yet the quoted (3800 +/- 1500) s error bar is only the spread of the GP samples. The systematic uncertainty from the GP method should be propagated into the reported lag and coronal-height estimates, or shown to be negligible for the specific 2013 observation geometry, before the quantitative value of the lag can be taken at face value.","section":"Section 4.1.2, Fig. 7; Section 5.3"}],"minor_comments":[{"comment":"There are numerous typographical errors (e.g., 'frequnecy', 'predicitons', 'requried', 'funtions', 'segnents', 'obervations') that should be corrected in a revised manuscript.","section":"Throughout"},{"comment":"The caption says 'a 500 s lag' while the text and the simulation description refer to a 5000 s lag; this should be harmonized.","section":"Fig. 7 caption"},{"comment":"The text states that the GP statistical uncertainty is 'around three times' that of continuous light curves, then later states that the Poisson-resampling test shows the GP uncertainty is 'around 10 times' that of continuous light curves; these statements should be reconciled and clarified.","section":"Section 4.1.2"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the manuscript is within MNRAS scope. The method section is a useful contribution and the simulations are carefully done. The main risk is that the reverberation interpretation is under-tested, but the identified gaps can be addressed within the manuscript's scope, so I do not see a basis for rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, the paper is worth taking seriously. The Gaussian-process machinery is not new, but the way it is used here — independent GP fits per energy band, then drawing sample light curves to build lag-frequency and lag-energy spectra across orbital gaps — is a real step beyond Zoghbi et al. 2013 and Reynolds 2000. The simulations are done properly: they test kernel choices, quantify coherence, show systematic offsets in some bins, and include a null test. The code is public. That part is solid.\n\nThe 3C 120 application is new and interesting: the first low-frequency iron K reverberation lag in a radio-loud AGN. The lag-energy profile has the expected shape, with the 5–7 keV bin lagging the 1.6–2 keV continuum by 3800 ± 1500 s, and the drop in the 7–10 keV bin does argue against a simple monotonic hard lag. The coronal height estimate of 13 ± 8 r_g is reasonable as a first guess.\n\nThe soft spots are exactly where the reader put them. The hard-lag alternative is dismissed in one sentence, without fitting a quantitative model. The error bars on the lag-energy points are sizable, and the GP systematic errors are frequency- and energy-dependent, so a smooth hard-lag curve could still be hidden inside the noise. The frequency band is chosen after looking at the data, and the 99.5% significance is an internal posterior-sample fraction, not a test against a null hypothesis. These don't kill the paper, but they mean the physical claim is provisional.\n\nThe paper is for the X-ray timing/reverberation subfield, and it deserves a serious referee. I would send it out with a request to fit a hard-lag model to the lag-energy spectrum, fold the measured systematics into the quoted lag, and state more carefully what is new versus what is inherited from Zoghbi et al. The framework itself is a useful resource.","headline":"Solid GP-based timing method with a plausible but not fully secured reverberation claim in 3C 120; worth refereeing with a requested hard-lag comparison.","tokens_in":25296,"tokens_out":2654,"would_cite":true,"duration_ms":25094,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Gaussian-process timing recovers a 3,800-second X-ray echo from the inner disc of the radio-loud AGN 3C 120.","keywords":["X-ray reverberation","Gaussian processes","active galactic nuclei","accretion disc","iron K line","time lags","3C 120","corona"],"falsifier":"Fit a quantitative hard-lag model in which the lag increases monotonically with energy to the measured $0.3$–$10$ keV lag-energy spectrum between $5\\times10^{-6}$ and $3\\times10^{-5}$ Hz; if such a model matches the data without an iron K peak, the reverberation interpretation is not unique. A longer observation that resolves the sharp drop above 7 keV and the rise below 1.6 keV would also test whether the claimed iron K line profile in the lag is real.","tokens_in":24238,"feed_emoji":"🔭","tokens_out":7747,"duration_ms":66177,"temperature":0.7,"pith_summary":"This paper claims that X-ray timing analysis can be pushed below the frequency floor set by orbital gaps by modelling each energy-band light curve as a Gaussian process, drawing continuous sample light curves, and computing Fourier-domain lags on those samples. It applies the method to two XMM-Newton orbits of the radio-loud AGN 3C 120 and reports a lag-energy spectrum between $5\\times10^{-6}$ and $3\\times10^{-5}$ Hz whose shape matches X-ray reverberation from the inner accretion disc. The 5–7 keV iron K line peak lags the 1.6–2 keV continuum-dominated band by $(3800\\pm1500)$ s, interpreted as light-travel delay between a corona and the disc, placing the corona at $(13\\pm8)\\,r_g$ above the disc. If the interpretation holds, this is the first reverberation measurement in a radio-loud AGN, and it opens low-frequency reverberation studies to low-Earth-orbit missions and higher-mass black holes.","feed_headline":"3C 120's disc echoes its X-rays 3,800 seconds late","feed_subtitle":"Gaussian-process timing bridges orbital gaps to catch the first disc reverberation lag in a radio-loud AGN.","key_machinery":"The central mechanism is a Gaussian process model of each energy-band light curve: the data are drawn from a multivariate Gaussian whose covariance is set by a kernel function, and after optimising the kernel hyperparameters, continuous light-curve realisations are drawn from the conditional distribution of unobserved times given the observed points. The paper uses the rational quadratic kernel, a mixture of squared-exponential correlations on many timescales, which reproduces the broken power-law power spectral density of AGN variability better than a single-scale kernel. It then computes the cross-spectrum, coherence, lag-frequency spectrum, and lag-energy spectrum on each pair of sample light curves and averages over thousands of samples, so the gaps are handled probabilistically rather than by interpolation.","core_discovery":"On its own terms, the paper establishes that a Gaussian process fitted independently to each X-ray energy band can preserve the phase relationship between bands across gaps, so that standard Fourier-domain lag measurements can be made at frequencies as low as $5\\times10^{-6}$ Hz from concatenated orbit segments. Applied to the 2013 XMM-Newton observations of 3C 120, the method yields a lag-frequency spectrum in which the 4–7 keV band lags the 1.2–4 keV band by $(6300\\pm2700)$ s at $7.5\\times10^{-6}$ Hz, and a lag-energy spectrum over $5\\times10^{-6}$ to $3\\times10^{-5}$ Hz with the characteristic iron K profile: the continuum-dominated 1.6–2 keV band responds earliest, the 5–7 keV core of the line responds latest, and the redshifted wing at 3–5 keV responds between them. The author interprets the $(3800\\pm1500)$ s lag between the 5–7 keV and 1.6–2 keV bands as X-ray reverberation from the inner disc, detected at 99.5 per cent confidence, and converts it to a coronal scale height of $(13\\pm8)\\,r_g$ above the disc.","pith_inferences":["A natural extension the paper does not pursue is applying the same Gaussian-process sampling procedure to covariance spectra and bispectra, which the paper notes are straightforward products of the same sample draws; this would let gap-ridden data probe non-linear variability and the log-normal flux distribution.","If the 86.3 per cent-confidence hint that the redshifted 3–4 keV wing does not lead the 1–2 keV continuum is confirmed by longer exposures, the lag-energy morphology would favour an extended disc-hugging corona over a slowly propagating jet-base fluctuation source in 3C 120.","The simulation results imply that at STROBE-X count rates the systematic bias near the orbital frequency (1/5700 s) becomes the limiting error, so a kernel matched more exactly to the power spectral density should be developed before the method is used on very high signal-to-noise data."],"forward_implications":["Low-frequency reverberation can be measured from light curves with gaps, so satellites in low-Earth orbit such as NICER, NuSTAR, and the proposed STROBE-X can perform reverberation experiments without continuous visibility.","Combining successive orbits extends reverberation studies to more massive AGN whose lags appear at lower Fourier frequencies than a single orbit can sample.","In 3C 120, the coronal height of $(13\\pm8)\\,r_g$ sits at the upper end of the iron K lag sample for Seyferts, suggesting that a jet-associated corona may be more extended than in radio-quiet AGN.","Comparing lag-energy profiles between radio-loud and radio-quiet AGN can constrain whether a collimated coronal core can coexist with a jet."],"supporting_citations":[{"why":"It supplies the earlier technique of fitting a cross-correlation model to unevenly sampled light curves that this method generalises by treating each band independently.","marker":"Zoghbi et al. 2013b"},{"why":"It provides the Gaussian process machinery, kernel formalism, and conditional distributions that the timing pipeline is built on.","marker":"Rasmussen & Williams 2006"},{"why":"It provides the algorithm used to simulate red-noise light curves with specified power spectra for the validation tests.","marker":"Timmer & Koenig 1995"},{"why":"It defines the standard Fourier-domain cross-spectrum and lag measurement into which the Gaussian process samples are fed.","marker":"Uttley et al. 2014"},{"why":"It supplies the spectral model of 3C 120 identifying the relativistically broadened iron K line from the inner disc and the jet contribution.","marker":"Lohfink et al. 2013"},{"why":"It provides the comparison sample of iron K lags in Seyfert galaxies and the mass scaling used to judge where 3C 120 sits.","marker":"Kara et al. 2016"},{"why":"It gives the black hole mass of 3C 120 used to convert the measured lag into a coronal height in units of $r_g$.","marker":"Pozo Nunez et al. 2012"},{"why":"It supplies the lag-to-height relation and the $1/(2\\tau)$ frequency limit used to predict where the reverberation signal should appear.","marker":"Wilkins & Fabian 2013"}],"fun_headline_variants":["First X-ray reverberation lag seen in radio-loud AGN 3C 120","Gaussian processes bridge gaps to find 3C 120's disc echo","Iron line lag in 3C 120 places corona at 13rg","New method uses Gaussian processes to spot disc reverberation","3C 120's iron K core echoes continuum 3800s later"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The identification of the lag-energy profile as disc reverberation rather than a continuum hard lag rests on its shape alone; the paper does not fit a quantitative hard-lag model to rule that alternative out.","fun_headline_variants_meta":{"raw":{"variants":["First X-ray reverberation lag seen in radio-loud AGN 3C 120","Gaussian processes bridge gaps to find 3C 120's disc echo","Iron line lag in 3C 120 places corona at 13rg","New method uses Gaussian processes to spot disc reverberation","3C 120's iron K core echoes continuum 3800s later"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00079,"raw_usage":{"total_tokens":3564,"prompt_tokens":1109,"completion_tokens":2455,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":725,"completion_tokens_details":{"reasoning_tokens":2356}},"tokens_in":725,"tokens_out":2455,"duration_ms":18152,"temperature":1.0,"reasoning_tokens":2356,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:56:16.113175+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit a quantitative hard-lag model in which the lag increases monotonically with energy to the measured $0.3$–$10$ keV lag-energy spectrum between $5\\times10^{-6}$ and $3\\times10^{-5}$ Hz; if such a model matches the data without an iron K peak, the reverberation interpretation is not unique. A longer observation that resolves the sharp drop above 7 keV and the rise below 1.6 keV would also test whether the claimed iron K line profile in the lag is real.","supporting_citations":[{"cited_title":"E., Williams C., 2006, Gaussian Processes for Machine Learning","cited_arxiv_id":null,"evidence_quote":"It provides the Gaussian process machinery, kernel formalism, and conditional distributions that the timing pipeline is built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the algorithm used to simulate red-noise light curves with specified power spectra for the validation tests."},{"cited_title":"M., Fabian A","cited_arxiv_id":null,"evidence_quote":"It defines the standard Fourier-domain cross-spectrum and lag measurement into which the Gaussian process samples are fed."}],"review_version":1}