{"id":"f712554a-0b3c-45e3-9a0c-80e93027178a","arxiv_id":"1908.10662","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"Gaussian process regression with celerite kernels models red-giant granulation and oscillations in the time domain, recovering nu_max to within about one percent, though some granulation parameters remain unconstrained.","lead":"This paper shows that Gaussian process models, fit with physically motivated kernels, can describe granulation and oscillation signals in red-giant light curves and recover the frequency of maximum oscillation amplitude to about one percent. It validates the approach on simulated TESS data and Kepler stars, and reports a preliminary improvement in transit radius measurements when stellar noise is modeled simultaneously.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported nu_max uncertainties sit below the paper's own Eq. (13) lower bound, so the central precision claim is unsupported without a coverage test.","rationale":"The accuracy of nu_max is well supported: the simulated recovery has about 1% bias, and the comparison with DIAMONDS shows about 1-2% relative agreement. The precision claim, however, is only as strong as the quoted uncertainties, and the paper's own Sec. 4.3 shows those uncertainties are below the stated lower bound. The appeal to DIAMONDS' literature success does not resolve the discrepancy, especially because DIAMONDS' intervals are even smaller. A direct coverage test on the simulated stars is cheap and decisive. I also noticed that Eq. (12) appears inconsistent with Eq. (A4) by a factor of sqrt(2), which would affect the physical amplitude mapping rather than nu_max; a typo correction would address that. The reader's conditional verdict is appropriate, and this concern adds a specific condition: calibrate or demonstrate the coverage of the reported nu_max uncertainties before relying on the precision claim.","tokens_in":15524,"tokens_out":19334,"duration_ms":209107,"concrete_test":"For the 20 simulated TESS-like stars with known nu_max, compute the empirical coverage of the 68.3% highest-posterior-density intervals reported by the GP fit across the 10 independent 27.4-day realizations. If fewer than roughly 68% of true nu_max values fall inside the quoted intervals, or if the RMS of per-realization nu_max estimates exceeds the mean reported sigma by more than about 20%, then the reported uncertainties understate the true precision and the claim of a precise nu_max measurement should be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim includes an accurate and precise measurement of nu_max. The accuracy part is supported by the simulations and by the 1-2% agreement with DIAMONDS. The precision part is not. In Sec. 4.3, Eq. (13) is introduced as a lower limit for the uncertainty in nu_max. The paper then reports, for the 19 Kepler stars, an average expected sigma of 7.15 microHz, an average GP sigma of 3.43 microHz, and an average DIAMONDS sigma of 1.91 microHz. The GP credible intervals are therefore only about half of the stated lower bound. The paper notices this but does not resolve it: the next sentence appeals to DIAMONDS' successful literature applications, which cannot establish that the GP intervals are correct. If Eq. (13) is a valid lower bound, the HPD intervals are systematically too small and the precise part of the central claim fails. If Eq. (13) is not applicable to a time-domain envelope centroid, then presenting it as a lower bound is misleading and the comparison in Sec. 4.3 is uninformative. Either way, the precision claim is not backed by any direct coverage check. This matters for downstream uses of the quoted uncertainties, such as scaling relations or ensemble asteroseismology.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a Gaussian-process (GP) framework, built on the celerite package, to model granulation and oscillation signals in red-giant light curves directly in the time domain. Two models are considered: Model 1 uses one granulation kernel, an oscillation-bump kernel, and white noise; Model 2 adds a second granulation component. The method is applied to 20 TESS-like artificial low-luminosity red-giant stars and to 19 Kepler LLRGB stars, and the recovered parameters are compared with the injected values and with DIAMONDS power-spectrum fits. The paper reports a 1.14% bias and 3.34% scatter in nu_max on the simulated data, relative biases of 1.99% and 0.92% against DIAMONDS for Models 1 and 2, and a preliminary improvement in Rp/Rstar when the GP stellar model is fit jointly with a transit model. The authors are transparent about limitations: the second granulation frequency is unconstrained for every simulated star, white noise is underestimated in several configurations, and the GP and DIAMONDS uncertainties for nu_max sit below the paper's own stated lower bound from Eq. (13).","tokens_in":15794,"tokens_out":7314,"duration_ms":74931,"significance":"If the main result survives revision, the paper makes a useful contribution: it provides a publicly available and physically motivated celerite-kernel model for red-giant stellar signals, and the nu_max accuracy is supported by simulation and by agreement with an independent power-spectrum method. I give credit to the public implementation, the reproducible workflow, and the unusually clear reporting of unconstrained parameters. The method is of direct practical relevance for TESS light curves and for joint transit-and-stellar-noise modelling. However, the stated precision of the nu_max measurement is not yet established, because the GP credible intervals are smaller than the stated lower bound from Eq. (13) and no coverage test is presented. The transit result is explicitly preliminary but is advertised in the abstract, which raises the bar for documenting that experiment. With these caveats, the paper's significance is moderate rather than high: the core ideas are sound, but the central precision claim needs either empirical calibration or reformulation.","major_comments":[{"comment":"The paper's central claim that nu_max can be measured both accurately and precisely is not fully supported. The authors call Eq. (13) a lower limit for the uncertainty in nu_max, yet report <sigma_GP> = 3.43 microHz and <sigma_DIAMONDS> = 1.91 microHz against <sigma_nu_max> = 7.15 microHz. The subsequent appeal to DIAMONDS' successful literature applications cannot validate the GP HPD intervals, since it provides no direct test of whether the 68.3% GP intervals contain the true nu_max in the simulations or in injected signals. I request either a coverage test on the TESS-like simulations with the empirical coverage probability reported, or a reformulation of the abstract and conclusions that claims accuracy only and presents the GP uncertainties as model-based estimates that may be underestimated. This matters for downstream uses of the quoted uncertainties, such as scaling relations or ensemble asteroseismology.","section":"Sec. 4.3, Eq. (13)"},{"comment":"The simulation validation is partially internal, because the artificial light curves are generated using the same Kallinger et al. (2014) Harvey-like granulation functional form that defines the GP kernel through Eqs. (9)-(12). The paper's own results show the limits of this mapping: the second granulation frequency b_gran,2 is unconstrained for every simulated star, b_gran,1 is recovered with a 8.05% bias in Model 1, and white noise is systematically underestimated in Model 2. Consequently, the claim of recovering physically meaningful parameters is convincingly established only for nu_max; it is not established for the individual granulation parameters. I request either a simulation test using a different background shape (e.g., a different exponent or an additional background component) or a revised conclusion that limits the physical-parameter claim to nu_max and the oscillation envelope.","section":"Sec. 3.2, Figs. 2-3"},{"comment":"The transit experiment is presented as a partial validation of the abstract's claim about improving Rp/Rstar, but the manuscript does not state the number of injected transits, the orbital periods or depths, the limb-darkening model, or the SDE threshold used for detection. Without these details the comparison in Fig. 10 cannot be reproduced or evaluated. If the transit claim is retained in the abstract, the experimental setup should be documented; alternatively, the sentence should be removed or moved to the context of future work.","section":"Sec. 5, Fig. 10"}],"minor_comments":[{"comment":"The text reports a relative white-noise bias of -48.71% with 16.89% scatter for Model 2, while the inset in Fig. 7 shows 1 sigma = 19.69%; this discrepancy should be reconciled.","section":"Sec. 4.2, Fig. 7"},{"comment":"The printed form of Eq. (13) is typographically ambiguous: the expression '(1 + 4 (H_BR/sigma_g)^{2/3})' could be read as a sum rather than as the expected ratio formula from Kallinger et al. (2010). Please verify the formula and clarify whether it is a strict lower bound or an empirical expectation, since the text relies on that distinction.","section":"Eq. (13)"},{"comment":"These figures show the PSD of the GP output and its components, but they do not overlay the input or DIAMONDS-comparable PSD model components; adding the true or reference background and oscillation envelope would help the reader judge how well the time-domain fit reproduces the frequency-domain structure.","section":"Fig. 5 and Fig. 9"},{"comment":"The authors correctly note that the Gelman-Rubin diagnostics are not strictly valid for correlated MCMC chains; however, the threshold of 1.1 and the practice of discarding up to 50% of each chain should be described with a reference or a brief justification, because these choices can affect the reported HPD intervals.","section":"Sec. 2.4"},{"comment":"The x-axis label 'Star' is uninformative; the figure would be clearer if the axis were labelled 'Star index (ordered by SDE)' and the SDE values were tabulated.","section":"Fig. 10"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest and well written, and the central accuracy result for nu_max is credible. The main block is the precision claim: the reported GP uncertainties are smaller than the paper's own lower bound, and no coverage test is presented. I think a revision that either adds coverage tests or removes the word 'precise' from the central claim would be sufficient, provided the transit claim is either documented properly or demoted to a clearly preliminary remark."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one with a specific question in mind: can a celerite GP kernel, with physically calibrated hyperparameters, do time-domain asteroseismology of red giants? The answer is yes for nu_max, with caveats. The genuinely new piece is the mapping between celerite SHOTerm parameters and the Kallinger/Harvey granulation parameters plus the oscillation bump (nu_max, a_gran, b_gran, P_g), and the validation on TESS-like simulations and a Kepler LLRGB sample. Prior GP applications (Barclay, Grunblatt) treated the GP as a noise filter; this paper tries to give the kernel parameters physical meaning, and mostly succeeds.\n\nWhat's done well: the paper is transparent about its limitations. It reports that the second granulation frequency cannot be constrained in the simulations, that white noise is underestimated, and that the best model depends on the data. The Kepler comparison against DIAMONDS is a sensible external benchmark, and nu_max agrees to 0.9–2%. The transit test is preliminary but sensible; the Rp/Rstar improvement is small and rightly described as preliminary. The code is public, and the citation pattern looks fine—prior GP red-giant work is cited.\n\nThe main soft spot is the uncertainty estimate for nu_max. The paper reports an average expected sigma from Kallinger's Eq. 13 of 7.15 microHz, but the GP gives 3.43 and DIAMONDS 1.91, roughly half or less of that lower bound. The paper acknowledges the discrepancy but then says DIAMONDS has been used successfully, which does not establish that the GP (or DIAMONDS) intervals are correctly calibrated. If Eq. 13 is a valid lower bound, both methods are overconfident; if it is not applicable to time-domain envelope centroid fits, presenting it as a lower bound is misleading. Either way, the precision claim needs a direct coverage test or a clear statement that the quoted uncertainties are relative to the model, not absolute. This is fixable but real.\n\nOne more caveat: the TESS simulations were generated with the same Harvey/Kallinger functional forms that the GP kernels approximate, so the simulation validates the fit within a model family, not the model family itself against real granulation. The Kepler comparison mitigates this somewhat but only tests consistency with another fitting method, not ground truth.\n\nOverall: this is a solid methods paper, worth refereeing seriously. I'd like to see the uncertainty issue addressed—a coverage test on the simulations would be the natural way—but I would not block publication if the authors soften the precise claim. Read it if you work on TESS red-giant transits or time-domain asteroseismology.","headline":"Solid, honest methods paper: the celerite-to-asteroseismic parameter mapping is genuinely new and nu_max recovery is robust, but the reported uncertainties need a calibration check before the precision claim is trusted.","tokens_in":16409,"tokens_out":3074,"would_cite":true,"duration_ms":30392,"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":"Gaussian processes can pull red-giant oscillation parameters straight out of light curves.","keywords":["asteroseismology","red giants","Gaussian processes","granulation","nu_max","celerite kernels","stellar noise","transit modelling"],"falsifier":"Fit the same two-component kernel to high-cadence, high-signal-to-noise Kepler light curves of red giants whose granulation background has been independently measured, and compare the recovered $a_\\mathrm{gran}$, $b_\\mathrm{gran}$, and $\\nu_\\mathrm{max}$ against flexible power-spectrum fits; systematic discrepancies beyond the few-percent biases reported here, or failure to recover an injected Harvey component with a different slope (for example exponent 2 instead of 4), would invalidate the claimed parameter mapping.","tokens_in":15317,"feed_emoji":"🌟","tokens_out":6856,"duration_ms":64513,"temperature":0.7,"pith_summary":"This paper argues that a Gaussian-process regression with a physically motivated kernel can model the granulation and solar-like oscillation signals of red-giant stars directly in the time domain, and can recover the frequency of maximum oscillation amplitude, $\\nu_\\mathrm{max}$, accurately. This matters because TESS will deliver on the order of $10^5$ red-giant light curves in which these stellar signals have amplitudes and timescales comparable to planetary transits, so a fast time-domain model with physical meaning can serve both asteroseismology and exoplanet transit fitting. The paper validates the claim on TESS-like artificial light curves and on 19 Kepler low-luminosity red giants, comparing recovered parameters with power-spectrum fits. It also reports a preliminary result that fitting the GP stellar model simultaneously with a transit model improves the precision and accuracy of the planet-to-star radius ratio, $R_p/R_\\star$.","feed_headline":"GP model pulls nu_max out of red-giant light curves","feed_subtitle":"A time-domain Gaussian-process fit recovers oscillation frequencies and sharpens planet-radius estimates for evolved stars.","key_machinery":"The central object is the celerite SHOTerm kernel, a covariance function whose power spectral density is that of a stochastically driven, damped harmonic oscillator. At quality factor $Q=1/\\sqrt{2}$ its PSD reduces to the Harvey-like form with an exponent of 4, the function usually used to model red-giant granulation in the frequency domain; for $Q>1$ the same kernel approximates the Gaussian-like shape of the oscillation bump. The model sums a granulation kernel (or two, in Model 2), an oscillation-bump kernel with $1.2<Q<18$, and a white-noise term. The kernel's hyperparameters $S_0$, $Q$, and $\\omega_0$ are then mapped one-to-one onto the asteroseismic parameters $a_\\mathrm{gran}$, $b_\\mathrm{gran}$, $P_g$, and $\\nu_\\mathrm{max}$ via the normalization relations derived in Appendix A, so the GP fit returns physical stellar parameters rather than merely a flexible noise model.","core_discovery":"The paper's central claim is that a Gaussian-process model built from damped-harmonic-oscillator kernels can represent the photometric signal of a red-giant star, granulation plus the solar-like oscillation bump, directly in the time domain, and that the fitted kernel parameters can be translated into the same physical quantities normally extracted by power-spectrum fitting. The translation is carried out through the mapping $a_\\mathrm{gran}=\\sqrt{\\sqrt{2}\\, S_{0,\\mathrm{gran}}\\,\\omega_{0,\\mathrm{gran}}}$, $b_\\mathrm{gran}=\\omega_{0,\\mathrm{gran}}/(2\\pi)$, $P_g=4S_{0,\\mathrm{bump}}Q^2$, and $\\nu_\\mathrm{max}=\\omega_{0,\\mathrm{bump}}/(2\\pi)$. On TESS-like simulated light curves the model recovers $\\nu_\\mathrm{max}$ with about 1% bias and 3-4% scatter, and on 19 Kepler low-luminosity red giants the time-domain and power-spectrum determinations of $\\nu_\\mathrm{max}$ agree to within about 1-2%. The paper also reports a preliminary combined fit of the GP stellar model with a transit model that reduces the bias in $R_p/R_\\star$ from 1.71% to -0.06% and the scatter from 6.49% to 5.80%. The authors note that a second granulation component's characteristic frequency is not constrained in the simulated data and that white noise is underestimated, which motivates their recommendation of the simpler one-granulation model for typical TESS time series.","pith_inferences":["Editorial inference: if the same kernel mapping holds at longer baselines and higher signal-to-noise, GP-based time-domain analysis could take over the role of periodogram-based asteroseismic fitting for large red-giant samples, avoiding binning choices and providing a single model that also handles transits; this is a plausible extension, not a paper claim.","Editorial inference: the cleanest test of the physical mapping is to let the granulation quality factor float in the fit; if the posterior converges near $Q=1/\\sqrt{2}$ across many Kepler giants, the one-to-one calibration is confirmed, whereas systematic deviation would mean the reported $a_\\mathrm{gran}$ and $b_\\mathrm{gran}$ carry model-dependent bias. Such a test is not reported in the paper.","Editorial inference: because the oscillation bump is treated as a single broad component, the recovered $\\nu_\\mathrm{max}$ describes the envelope rather than individual mode frequencies; combining this time-domain model with a mode-by-mode search could produce a pipeline that first extracts stellar parameters and then measures frequency spacings from the same framework."],"forward_implications":["For 27.4-day TESS-like light curves of red giants, the simpler one-granulation model recovers $\\nu_\\mathrm{max}$ with roughly 1% bias and 3-4% scatter on simulated data, meaning asteroseismic characterization can be done in the time domain without building a full frequency-domain model.","On real Kepler low-luminosity red giants, the time-domain GP and power-spectrum fits agree in $\\nu_\\mathrm{max}$ to within about 1-2%, indicating that the kernel choice does not introduce large systematic offsets in the main quantity of interest.","Modelling the stellar GP simultaneously with a transit improves the estimated $R_p/R_\\star$ from a 1.71% offset with 6.49% scatter to a -0.06% offset with 5.80% scatter in simulated injections, so evolved-host transit parameters benefit directly from treating stellar signals as correlated noise.","Because the celerite kernels allow linear-scaling likelihood evaluation, the method can be applied to long TESS light curves, and the implementation is publicly available.","The second granulation component's frequency is left unconstrained by TESS-like data, and white-noise levels are systematically underestimated in the two-component model, so the paper recommends Model 1 for typical TESS red-giant light curves."],"supporting_citations":[{"why":"Supplies the celerite package, the SHOTerm kernel, and the damped-harmonic-oscillator PSD on which the entire parameter mapping rests.","marker":"Foreman-Mackey et al. 2017"},{"why":"Defines the two-component Harvey-like granulation background and the power-spectrum parameters $a_\\mathrm{gran}$, $b_\\mathrm{gran}$, $P_g$, and $\\nu_\\mathrm{max}$ to which the GP parameters are compared.","marker":"Kallinger et al. 2014"},{"why":"Provides the exponent-4 granulation form for red giants and the expected $\\nu_\\mathrm{max}$ uncertainty relation used in Section 4.3.","marker":"Kallinger et al. 2010"},{"why":"Provides the DIAMONDS power-spectrum fitting code that serves as the baseline comparison for the Kepler star analyses.","marker":"Corsaro & De Ridder 2014"},{"why":"Supplies the sample of 19 Kepler low-luminosity red-giant stars and the three-component background discussion motivating Model 2.","marker":"Corsaro et al. 2015"},{"why":"Provides the TESS photometric noise model and the oscillation-bump width scaling relation used in the simulated data and the uncertainty estimate.","marker":"Campante et al. 2016"},{"why":"Supplies the transit model that is combined with the GP stellar model in Section 5.","marker":"Kreidberg 2015"},{"why":"Provides the TESS noise model used in generating the artificial light curves.","marker":"Sullivan et al. 2015"}],"fun_headline_variants":["GP fit digs red-giant oscillation peak from noise","Time-domain GP recovers red-giant oscillation frequency","GP model sharpens planet radii around red giants","Red-giant noise tamed by GP aids exoplanet measurements","Time-domain GP nails red-giant oscillation peak"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the assumption that the chosen kernel's power spectral density, a damped harmonic oscillator with quality factor fixed at $1/\\sqrt{2}$ for granulation, matches the real granulation signal of red giants closely enough that the fitted kernel hyperparameters correspond one-to-one with the asteroseismic amplitudes, frequencies, and $\\nu_\\mathrm{max}$.","fun_headline_variants_meta":{"raw":{"variants":["GP fit digs red-giant oscillation peak from noise","Time-domain GP recovers red-giant oscillation frequency","GP model sharpens planet radii around red giants","Red-giant noise tamed by GP aids exoplanet measurements","Time-domain GP nails red-giant oscillation peak"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001203,"raw_usage":{"total_tokens":5068,"prompt_tokens":1169,"completion_tokens":3899,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":785,"completion_tokens_details":{"reasoning_tokens":3822}},"tokens_in":785,"tokens_out":3899,"duration_ms":31243,"temperature":1.0,"reasoning_tokens":3822,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:37:45.176898+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit the same two-component kernel to high-cadence, high-signal-to-noise Kepler light curves of red giants whose granulation background has been independently measured, and compare the recovered $a_\\mathrm{gran}$, $b_\\mathrm{gran}$, and $\\nu_\\mathrm{max}$ against flexible power-spectrum fits; systematic discrepancies beyond the few-percent biases reported here, or failure to recover an injected Harvey component with a different slope (for example exponent 2 instead of 4), would invalidate the claimed parameter mapping.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the exponent-4 granulation form for red giants and the expected $\\nu_\\mathrm{max}$ uncertainty relation used in Section 4.3."}],"review_version":1}