{"id":"6c255918-954f-4000-b4df-080f52207f33","arxiv_id":"2504.12420","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"gwharmone is a data-driven surrogate that reproduces the eccentric harmonics of the dominant quadrupole mode in non-spinning eccentric binary black hole waveforms with average frequency-domain mismatches near 0.004.","lead":"Researchers built a fast computer model, gwharmone, that predicts the four eccentric harmonics in gravitational waves from black hole mergers on elliptical orbits. It runs in about 0.1 seconds, roughly a thousand times faster than the full calculation, which could help future searches for these rare signals.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The mean-anomaly dependence is asserted via an unpublished companion relation and receives no validation at lref != pi, so the model's headline capability is not yet demonstrated.","rationale":"Readers should take the paper at its word: the object is a surrogate for TEOBResumS eccentric harmonics, not for full NR. Within that frame, the interpolation machinery is standard and the lref=pi validation is reasonable: 162 training points, 1751 validation points, GPR with SVD bases, and reconstruction-error baseline are all standard practice in surrogate modeling. The decisive gap is that the abstract and Model evaluation promise lref as an input, yet the training data and the validation pipeline are fixed at lref=pi, with lref variation handled by an unpublished relation. That makes the correctness of this relation load-bearing: if it is wrong, the model silently fails for most of the input space it advertises, regardless of the excellent lref=pi errors. The j=1/j=3 phase reconstruction has the same provenance, but it is exercised by the lref=pi validation, so it is less fragile. We therefore keep the reader's CONDITIONAL verdict: likely sound for the fixed-mean-anomaly case, but the headline capability requires the companion paper and lref-dependent validation. This is not a rejection because no internal inconsistency or obviously wrong step was found; it is a request for missing evidence.","tokens_in":13766,"tokens_out":7265,"duration_ms":76832,"concrete_test":"Generate TEOBResumS+gwMiner full-order waveforms at 20 (q,e_ref) points spanning the target space, for lref in {0, pi/2, pi, 3pi/2, 2pi} (with e_ref defined at the same reference time/frequency as in training), and compare gwharmone predictions to them using the same time-domain L2 error and Advanced-LIGO mismatch definitions as Figures 2-3. If the lref != pi mismatches are comparable to the lref = pi values, the mean-anomaly claim is supported; if they are systematically worse (e.g., exceed 3% or degrade the Figure 3 distribution), the companion relation is insufficient. Independently, make Ref. [75] public (or include Eq. (24) and the SVD-coefficient-to-lref mapping in this paper) so the relation can be checked by re-derivation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Training data are generated only at lref = pi (Snapshot data section: 'we fix mean anomaly lref = pi'), and the lref direction is then inserted through a 'phenomenologically observed universal relation between eccentric harmonics at different lref values' and, in Model evaluation, a 'universal relation between SVD coefficients and mean anomaly values,' both deferred to Ref. [75], listed as in preparation. No equation for this relation appears in the manuscript, and no lref-dependent validation is reported: the 1751 validation points are described only as 'randomly distributed throughout the parameter space,' while all harmonic extraction is fixed at lref = pi and the accuracy comparisons (Figures 2-3) never state that lref is varied. If the validation is at lref = pi, then the claimed dependence on mean anomaly has not been tested at all, and an error or restricted validity of the companion relation would leave every reported number unchanged while breaking the model for lref != pi. The j=1 and j=3 phase relations (Eqs. 10-11, from Eq. (24) of Ref. [75]) are also imported rather than derived, but because the lref = pi validation waveforms contain these harmonics, the phase relations are at least empirically checked at lref = pi. The unvalidated lref map is therefore the single load-bearing step; the lref = pi surrogate may be sound, but the central claim as stated is conditional on an unavailable source.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents gwharmone, a surrogate model trained on TEOBResumS eccentric waveforms that predicts the first four eccentric harmonics (j=1,...,4) of the dominant quadrupolar mode for non-spinning binary black hole mergers, together with the full (2,2) mode. The model uses SVD-based model order reduction and Gaussian process regression over mass ratio q in [1,3.5] and reference eccentricity e_ref in approximately [0.03,0.175], with training data generated at a fixed mean anomaly l_ref=pi. The claimed l_ref dependence is inserted through a 'universal relation' imported from a companion paper (Ref. [75], in preparation), as are the phase relations for the j=1 and j=3 harmonics. Reported validation errors are around 1e-3 in the time domain and average mismatches around 0.004 against Advanced LIGO noise, with evaluation cost around 0.1 seconds.","tokens_in":14157,"tokens_out":4220,"duration_ms":44784,"significance":"If the claims are substantiated, gwharmone would be a practically useful fast surrogate for TEOBResumS eccentric waveforms, with clear value for searches and parameter estimation. The paper's strengths include a clear hierarchical modeling strategy, a concrete SVD/GPR pipeline, validation against the same approximant used for training (the appropriate test for a surrogate), and a commitment to public release through gwModels. The central limitation is that the mean-anomaly direction, which is part of the headline claim, is not trained on data but is imported from an unpublished companion paper and is not validated at l_ref != pi. The j=1 and j=3 phase relations are likewise imported and are not validated per harmonic. These issues are fixable within the manuscript's scope, but they currently leave the central claim conditional.","major_comments":[{"comment":"The mean-anomaly direction is the only part of the model that is not trained on data. All 162 eccentric training points are generated at l_ref=pi, and the l_ref dependence is inserted through a 'phenomenologically observed universal relation between eccentric harmonics at different lref values' and a 'universal relation between SVD coefficients and mean anomaly values,' both attributed to the unpublished Ref. [75]. No equation for either relation is given, and the accuracy section does not state that l_ref is varied in the 1751 validation points. If those points are all at l_ref=pi, every reported error is consistent with an incorrect or restricted l_ref relation. Because the abstract explicitly claims that the model 'includes the effect of mean anomaly,' this step must be made available (or derived) and validated with l_ref varied, or the claim of mean-anomaly dependence must be withdrawn.","section":"Snapshot data, Model evaluation, Model accuracy"},{"comment":"The phases of the j=1 and j=3 harmonics are imposed via phi_22,1 = phi_lambda + phi_ecc and phi_22,3 = 3 phi_lambda + phi_ecc + pi, taken from Eq. (24) of Ref. [75], a companion paper listed as in preparation. These relations are not derived in the manuscript, and the reported accuracy metrics (Fig. 2 and Fig. 3) are for the full h22 waveform, not for each harmonic separately. The fidelity of the individual predicted harmonic components is therefore not directly demonstrated. Adding per-harmonic validation that compares each h22,j(t) against the gwMiner extraction at held-out points would substantiate the 'surrogate for eccentric harmonics' claim.","section":"Model evaluation, Eqs. (10)-(11)"},{"comment":"The 1751 validation points are described only as 'randomly distributed throughout the parameter space.' The parameter ranges for q, e_ref, and l_ref over which these points are drawn are not specified. This is important not only for interpreting the average error but also for determining whether the mean-anomaly direction has been tested at all. Please state the validation distribution explicitly, including the l_ref values, and report mismatches binned in l_ref.","section":"Model accuracy"}],"minor_comments":[{"comment":"The SVD dimensions are stated incorrectly: for a data matrix X of shape n x m, the factorization is X = U Sigma V^T with U of shape n x n, Sigma of shape n x m, and V^T of shape m x m, not U of shape m x m and V^T of shape m x n as written.","section":"Model order reduction"},{"comment":"The abstract states the training eccentricity range as e_ref in [0,0.2] while the target parameter space is defined later as e_ref in [0.03,0.175] with training extending to 0.2. Please make this distinction explicit at first mention to avoid confusion between the training range and the validated target range.","section":"Abstract and Summary"},{"comment":"The evaluation-time comparison is computed at the training parameter points only. A few timing measurements at validation points, or at least a statement that the cost is parameter-independent, would make the speed comparison more robust.","section":"Figure 4 and Model evaluation cost"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is heavily dependent on a companion paper, Ref. [75], which is listed as in preparation and provides both the mean-anomaly map and the j=1/j=3 phase relations. I recommend that the editor require the authors either to include the necessary derivations or equations in this manuscript or to make the companion paper available during review. The l_ref=pi surrogate may be sound, but the paper's headline claim of mean-anomaly dependence is currently not testable by the reader."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this if you care about eccentric waveform surrogates. The paper delivers the first data-driven surrogate of eccentric harmonics, not just the full waveform. Architecture is sensible: circular base model, then eccentric corrections, with SVD + GPR. The speed gain (0.1 s vs ~300 s for the full-order extraction) is real, and the validation errors against TEOBResumS (~10^-3 time-domain, 0.004 mismatch) are plausible for a surrogate.\n\nThe soft spot is the mean-anomaly dependence. Training data are generated only at lref = pi. The lref direction is then inserted via a 'universal relation' from the companion paper (Ref [75], in preparation). No equation for that relation appears in the manuscript, and I find no evidence that the 1751 validation points vary lref. The claim 'includes the effect of mean anomaly' is therefore untested. If that relation is wrong or narrower than the parameter range, the model's output at lref != pi is wrong while every reported number stays the same. That's a load-bearing gap, not a cosmetic one.\n\nThe j=1 and j=3 phase relations are also imported from the companion, but those are indirectly checked at lref = pi because the summed harmonics are compared to TEOBResumS. So the fixed-lref surrogate itself looks sound. Also note: code is not yet public (gwMiner on request, gwModels forthcoming), and the model is limited to non-spinning, moderate eccentricity, and only the (2,2) mode harmonics. That's a reasonable scope for a first paper.\n\nRecommendation: send to peer review. The authors should be asked to either validate the lref direction with withheld data or explicitly restrict the claim to lref = pi. With that fixed, this is a useful contribution for search and PE efforts targeting eccentric binaries.","headline":"Solid surrogate for eccentric harmonics at fixed mean anomaly, but the headline mean-anomaly capability is imported from an unpublished companion and never validated.","tokens_in":14618,"tokens_out":5104,"would_cite":true,"duration_ms":49734,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["04.30.-w"],"model":"deepseek-v4-flash","headline":"gwharmone reproduces eccentric merger harmonics in 0.1 seconds","keywords":["gravitational waves","eccentric binary black holes","surrogate model","eccentric harmonics","Gaussian process regression","singular value decomposition","effective-one-body waveforms","mean anomaly"],"falsifier":"Run the harmonic-extraction pipeline on effective-one-body waveforms at parameter points inside the training box with mean anomalies far from $\\pi$ (for example $l_{\\rm ref}=0$ or $\\pi/2$), and compare the $j=1$ and $j=3$ phases and the full $(2,2)$ mode against gwharmone; mismatches well above the reported one percent would show the universal mean-anomaly or phase relations are not correct. A second check is to compare gwharmone against eccentric numerical-relativity waveforms in the same mass-ratio and eccentricity range, which would reveal whether the surrogate has learned the physics or only the approximant.","tokens_in":13596,"feed_emoji":"🌊","tokens_out":11992,"duration_ms":102045,"temperature":0.7,"pith_summary":"This paper presents gwharmone, the first data-driven surrogate model for the eccentric harmonics that make up the dominant quadrupolar gravitational-wave mode of eccentric, non-spinning binary black hole mergers. The authors try to show that a model trained on 173 effective-one-body waveforms can rapidly predict the four leading harmonics and the full quadrupolar mode for mass ratios $q\\in[1,3.5]$ and eccentricities $e_{\\rm ref}\\in[0.03,0.175]$, including dependence on the mean anomaly, with frequency-domain mismatches below one percent against the training approximant. If true, this removes a major computational bottleneck: the full harmonic extraction takes about 300 seconds per waveform, while the surrogate evaluates in about 0.1 seconds. That speed matters because measuring eccentricity in detected mergers is a promising route to identifying how binary black holes form.","feed_headline":"gwharmone reproduces eccentric merger harmonics in 0.1 seconds","feed_subtitle":"A 0.1-second evaluation makes eccentric mergers practical for searches and parameter estimation.","key_machinery":"The central object is the eccentric harmonic decomposition: each spherical harmonic mode $h_{\\ell m}(t)$ is split into a sum of monotonic harmonics $h_{\\ell m,j}(t)$ whose frequencies and phases follow a hierarchical structure. The argument runs on the hierarchical surrogate built from this decomposition: a circular base model for amplitude and phase, delta amplitude and phase corrections for the dominant $j=2$ harmonic, direct models for the $j=1,3,4$ amplitudes, and phase relations $\\phi_{22,1}=\\phi_\\lambda+\\phi_{\\rm ecc}$ and $\\phi_{22,3}=3\\phi_\\lambda+\\phi_{\\rm ecc}+\\pi$ that reconstruct the odd harmonics from the orbital and eccentric phases derived from $j=2$ and $j=4$. SVD provides the reduced basis and Gaussian process regression interpolates the coefficients; together these carry the speed and smoothness claims.","core_discovery":"The paper's central claim is that the eccentric harmonic decomposition of a gravitational-wave mode, not the full oscillatory waveform, is the right object to model. Each spherical harmonic mode is written as a sum of monotonic harmonics $h_{\\ell m}(t)=\\sum_j h_{\\ell m,j}(t)$, and the surrogate models the amplitudes and phases of the first four harmonics $j=1,2,3,4$ separately. The $j=2$ harmonic is treated as the circular waveform plus eccentric corrections; the $j=4$ harmonic gets its own correction; and the $j=1$ and $j=3$ phases are reconstructed from phenomenologically derived phase relations from the companion paper. Singular value decomposition compresses each data piece to one to three basis vectors, Gaussian process regression interpolates the coefficients in $(q,e_{\\rm ref})$, and a claimed universal relation supplies the mean-anomaly direction. The result, the paper argues, is a model whose validation errors are around $10^{-3}$ in the time domain, whose average advanced-LIGO mismatch is about 0.004, and whose evaluation cost is roughly 0.1 seconds.","pith_inferences":["The strongest test of the model is not the reported validation against the same approximant, but a comparison at mean anomalies far from the training value $l_{\\rm ref}=\\pi$ and against independent numerical-relativity eccentric waveforms.","If the universal mean-anomaly relation from the companion paper is confirmed, the same trick could be reused for other waveform approximants, avoiding the expensive per-point harmonic extraction grid.","The monotonicity of eccentric harmonics may enable parameter-estimation speed-ups beyond searches, such as heterodyned likelihood evaluations that exploit smoothly evolving phases.","Users should treat the claimed $l_{\\rm ref}$ coverage as provisional until the companion paper appears, since that direction is supplied by asserted relations rather than trained data."],"forward_implications":["A template bank for eccentric binary black hole searches becomes affordable: one model evaluation costs about 0.1 seconds rather than the roughly 300 seconds of a full harmonic extraction.","The model returns individual monotonic eccentric harmonics, so mode-by-mode filtering and fast likelihood schemes can be built directly on its output.","Because the $j=2$ harmonic is modeled as circular plus corrections, the same machinery doubles as a fast surrogate for the full quadrupolar mode of the effective-one-body approximant.","The stated accuracy holds throughout $q\\in[1,3.5]$ and $e_{\\rm ref}\\in[0.03,0.175]$, with degradation only near the high-eccentricity edge, so the model is usable across most of its advertised box.","The modeling framework is stated to carry over to eccentric, non-precessing systems with spins, making the harmonic representation a plausible basis for future eccentric waveform models."],"supporting_citations":[{"why":"Supplies the harmonic-extraction framework, the universal relation linking harmonics across mean-anomaly values, and the phase relations used to reconstruct the j=1 and j=3 harmonics.","marker":"[75]"},{"why":"The effective-one-body eccentric approximant that generated the 173 training waveforms and all validation waveforms.","marker":"[38]"},{"why":"Introduced the harmonic decomposition of eccentric binary waveforms that the surrogate models.","marker":"[74]"},{"why":"An earlier eccentric-binary surrogate whose data-driven methods and boundary-error behavior are reused here.","marker":"[49]"},{"why":"A surrogate-model construction using SVD reduced bases and regression, the technique that carries gwharmone's speed.","marker":"[76]"},{"why":"Defines the time-domain, time-and-phase-optimized relative L2 error used to report training and validation errors.","marker":"[82]"},{"why":"Defines the mismatch used for the frequency-domain accuracy claims against the advanced LIGO noise curve.","marker":"[83]"},{"why":"Provides the advanced LIGO sensitivity curve used in the mismatch calculation.","marker":"[84]"}],"fun_headline_variants":["First surrogate for eccentric merger harmonics: 0.1s","gwharmone: eccentric harmonics in 0.1s","Data-driven eccentric harmonics: 0.1s and ~0.001 error","Eccentric BBH harmonics: surrogate cuts evaluation to 0.1s","New eccentric surrogate: 0.1s, sub-0.01 mismatch"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the companion paper's universal relations for how the harmonics change with the mean anomaly (the orbital angle at the reference time) and for the phases of the $j=1$ and $j=3$ harmonics are correct, since the surrogate trains at only one mean-anomaly value and uses those asserted relations for every other value.","fun_headline_variants_meta":{"raw":{"variants":["First surrogate for eccentric merger harmonics: 0.1s","gwharmone: eccentric harmonics in 0.1s","Data-driven eccentric harmonics: 0.1s and ~0.001 error","Eccentric BBH harmonics: surrogate cuts evaluation to 0.1s","New eccentric surrogate: 0.1s, sub-0.01 mismatch"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000754,"raw_usage":{"total_tokens":3397,"prompt_tokens":1030,"completion_tokens":2367,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":646,"completion_tokens_details":{"reasoning_tokens":2268}},"tokens_in":646,"tokens_out":2367,"duration_ms":18076,"temperature":1.0,"reasoning_tokens":2268,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:31:56.859935+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the harmonic-extraction pipeline on effective-one-body waveforms at parameter points inside the training box with mean anomalies far from $\\pi$ (for example $l_{\\rm ref}=0$ or $\\pi/2$), and compare the $j=1$ and $j=3$ phases and the full $(2,2)$ mode against gwharmone; mismatches well above the reported one percent would show the universal mean-anomaly or phase relations are not correct. A second check is to compare gwharmone against eccentric numerical-relativity waveforms in the same mass-ratio and eccentricity range, which would reveal whether the surrogate has learned the physics or only the approximant.","supporting_citations":[],"review_version":1}