{"id":"20e1624d-15c7-4d5f-80b9-c3e9eb481211","arxiv_id":"2508.06419","paper_version":1,"verdict":"UNVERDICTED","confidence":"UNKNOWN","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"Oxalate spectra and proton dynamics were reportedly simulated with a CCSD(T)-quality machine-learned PES predicting a 35 cm^-1 tunneling splitting, but the supplied text is a different paper.","lead":"This preprint claims a machine-learned model of protonated oxalate that reproduces new infrared spectra and predicts a 35.0 cm^-1 proton tunneling splitting. The enclosed full text is an unrelated heat-transfer simulation, so the chemistry claims could not be audited from this submission.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The abstract's validation claim rests on an unseen train/test split: 'calibrated against new spectroscopic measurements' may include the 1666 cm^-1 and H-transfer target bands, and the supplied full text contains no oxalate content to check this.","rationale":"The reader's UNVERDICTED assessment is correct: the supplied full text is an unrelated CFD manuscript, so none of the oxalate claims can be audited. I partially agree with the reader's identified weakest assumption: the calibration-versus-validation ambiguity is the sharpest conceptual soft spot in the abstract, independent of the document mismatch. The full-text mismatch is decisive for unverdictability, but even with a matching full text, the 'correctly predicted 1666 cm^-1 signature' would only be convincing if that band and the H-transfer region were excluded from PES calibration. My proposed check directly tests that separation and also demands a reproducible instanton calculation. I do not see grounds to ACCEPT or REJECT; the evidence is absent, so the verdict should remain UNVERDICTED.","tokens_in":5208,"tokens_out":2750,"duration_ms":30404,"concrete_test":"Retrieve the actual oxalate manuscript and its SI. Check the ML-PES fitting procedure for a list of spectroscopic targets included in the training set. If the 1666 cm^-1 signature or the 2940 cm^-1 H-transfer band was included, retrain (or, if code is unavailable, have the authors retrain) the PES with those bands withheld and recompute VPT2 frequencies; if the predicted bands shift by more than ~5–10 cm^-1, the claimed validation is circular. Separately, if the PES is released, run an independent ring-polymer instanton calculation with the stated higher-order corrections and confirm Δ_H = 35.0 cm^-1; if no PES/code is released, mark the tunneling prediction as unverifiable.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The central claim—that the ML-PES is quantitatively validated by new experiment and yields a falsifiable tunneling benchmark (Δ_H = 35.0 cm^-1)—requires two conditions: (i) the target observables used for comparison (the 1666 cm^-1 band, the 2940 cm^-1 H-transfer feature, framework modes) were not used to fit the PES, otherwise the agreement is a property of the fit; and (ii) the computational pipeline (VPT2, MD, ring-polymer instanton with higher-order corrections) can be audited against the PES. The abstract states only that the PES was 'calibrated against the results of new spectroscopic measurements' and never states an exclusion criterion. The supplied full text is an OpenFOAM CFD study of natural convection in vertical channels (References [1]-[35]); it contains no oxalate PES, no VPT2/MD/instanton results, no equations or SI for the claimed calculations. Thus the strongest claim cannot be checked at all in this document, and the abstract's wording raises a concrete circularity risk: if the 1666 cm^-1 signature and the 2600–3200 cm^-1 H-transfer region were in the training data, the 'correctly predicted' band is not independent evidence, and the tunneling splitting inherits any deficiencies of the fitted surface along the transfer path.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript consists of an abstract describing a machine-learned potential energy surface (PES) for protonated oxalate at CCSD(T) quality, VPT2 and molecular dynamics infrared assignments, ring-polymer instanton tunneling splittings, and new infrared spectroscopic measurements. The abstract reports that the PES was 'calibrated against' the new spectra, that VPT2 'correctly predicted' a new 1666 cm^-1 band, and that the H-transfer tunneling splitting is predicted as Δ_H = 35.0 cm^-1. The full text supplied after the abstract, however, is an OpenFOAM CFD study of laminar natural convection in open-ended vertical channels: it contains no oxalate PES, no electronic-structure calculations, no VPT2/MD/instanton equations, no experimental infrared data, and no supplementary material. The references [1]–[35] are all CFD/heat-transfer references. Thus the manuscript's central claims are entirely unsupported by any presented evidence.","tokens_in":5448,"tokens_out":2659,"duration_ms":32456,"significance":"If the abstract's claims were substantiated, the work could be significant: a machine-learned PES for a small molecular ion, benchmarked against new spectroscopy and yielding a falsifiable tunneling splitting, would be a useful contribution to chemical physics. The claimed validation of a CCSD(T)-quality PES against new experimental bands, and the prediction of Δ_H = 35.0 cm^-1, would be of interest to both experimental and theoretical spectroscopists. However, the manuscript as submitted provides no derivations, data, convergence tests, or error bars. The only oxalate-specific content is the abstract; the body is an unrelated CFD study. No machine-checkable proof, reproducible code, or parameter-free derivation is present. The potential circularity from calibrating the PES against the same measurements used for validation is an additional concern that cannot be resolved from the supplied text. As a result, the significance of the reported results cannot currently be assessed.","major_comments":[{"comment":"The body of the manuscript is an OpenFOAM CFD study of natural convection in vertical channels and contains no oxalate content. There is no description of the machine-learned PES, its training set, descriptors, or fitting procedure; no VPT2, molecular dynamics, or ring-polymer instanton equations; no experimental infrared spectra; and no data tables or error bars. Every claim in the abstract — the PES calibration, the 1666 cm^-1 prediction, the 2940 cm^-1 H-transfer assignment, and Δ_H = 35.0 cm^-1 — is unsupported. This is not a local omission but a wholesale absence of the paper's subject matter.","section":"Full text (entire manuscript after the abstract)"},{"comment":"The abstract states that the PES was 'calibrated against the results of new spectroscopic measurements' and later cites the VPT2 calculation's 'correctly predicted' 1666 cm^-1 band and agreement with measured framework and H-transfer modes as success. No exclusion criterion is given. If the 1666 cm^-1 band or the 2600–3200 cm^-1 H-transfer region were part of the calibration data, the agreement is a property of the fit, not independent validation, and the tunneling splitting inherits the quality of the fitted surface along the transfer path. The manuscript must specify which experimental data were used in fitting and which were held out for validation.","section":"Abstract"},{"comment":"No computational details are provided for the claimed VPT2, molecular dynamics, or ring-polymer instanton calculations with higher-order corrections. There are no basis sets, electronic-structure reference levels, convergence tests, or uncertainty estimates. Without these, the central prediction Δ_H = 35.0 cm^-1 is not auditable. The provided text also contains no experimental methods for the new spectroscopic measurements, so the claimed benchmark cannot be evaluated.","section":"Full text (methods and results)"},{"comment":"The reference list is entirely devoted to natural convection and heat-transfer CFD topics (e.g., Desrayaud et al., OpenFOAM solvers). There are no citations to the oxalate literature, machine-learned PES methods, VPT2 theory, ring-polymer instanton methodology, or infrared spectroscopy of carboxylic acids. This confirms that the full text is a different manuscript and not an incomplete version of the oxalate paper.","section":"References [1]–[35]"}],"minor_comments":[{"comment":"The phrase 'calibrated against the results of new spectroscopic measurements' should be replaced with a precise statement of the train/test split. If the target bands were excluded from fitting, that should be stated explicitly; if not, the validation claims should be reframed.","section":"Abstract"},{"comment":"The CFD equations in the body are garbled (e.g., Eq. (2) is missing a brace, and several boundary conditions appear as corrupted encoding). While this is secondary to the main problem, it further indicates that the manuscript is not in a reviewable form.","section":"Full text (equations)"}],"recommendation":"reject","confidential_remarks":"The supplied manuscript appears to have an abstract for one paper and a full text for a completely different paper. This is not a conventional revision issue; the entire scientific content claimed in the abstract is missing. The editor may wish to verify the submission integrity. If this is a file-upload error, a corrected manuscript could be re-reviewed, but as submitted, the paper cannot be accepted or meaningfully revised in its present form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — I looked at arXiv:2508.06419. The headline: the abstract announces a plausible ML-PES study of protonated oxalate, but the full text I received is an OpenFOAM CFD study of natural convection in vertical channels. Those two documents are not the same paper, and that mismatch is decisive for any editorial decision. There is no oxalate PES, no VPT2/MD/instanton results, no equations, no SI to audit here.\n\nLet me give the abstract its due. The scientific plan is coherent and, if executed as summarized, useful: build a CCSD(T)-quality machine-learned surface, calibrate it against new spectra, use VPT2 and MD to assign framework and H-transfer modes, and then compute the H-transfer tunneling splitting with ring-polymer instanton. The specific claims — a new low-intensity band at 1666 cm^-1 correctly predicted, a 2940 cm^-1 H-transfer assignment on a broad background, and a Δ_H = 35.0 cm^-1 benchmark — are concrete and falsifiable. This is a solid application of established methods to a prototypical carboxylate system, and the tunneling splitting as a benchmark for PES quality is a good idea.\n\nThe soft spots are serious. First, the circularity worry is real and visible in the abstract: the PES is 'calibrated against the results of new spectroscopic measurements,' and then the VPT2 agreement with those same measurements, including the 1666 cm^-1 band, is reported as success. Unless the target bands were held out of the calibration, this is fitting, not validation, and the tunneling splitting inherits the quality of that fit. The abstract never states an exclusion criterion. Second, the provided full text is unrelated to the abstract, so no derivation, convergence test, or error bar can be checked. Even a generous referee has nothing to evaluate. Third, the abstract alone lacks training set composition, basis set details, instanton convergence, and uncertainty estimates — all presumably in the real paper, but absent here.\n\nI don't think this is a case where the author is sloppy in a minor way. As submitted, the document is internally inconsistent. A serious editor would desk reject it and ask the authors to resubmit with the correct manuscript attached. If the mismatch is a repository error, that resubmission will fix it. If the abstract is what the authors actually intend to publish, they also need to address the calibration-versus-validation issue before it goes to referees.\n\nRecommendation: desk reject. Not because the underlying research, as advertised, is unpromising — it isn't — but because this submission cannot be reviewed. Ask for a corrected version.","headline":"Abstract describes a plausible and potentially valuable ML-PES study of protonated oxalate, but the supplied full text is an unrelated OpenFOAM CFD paper, so the submission as-is cannot be refereed.","tokens_in":6053,"tokens_out":3025,"would_cite":false,"duration_ms":30604,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A machine-learned potential energy surface of CCSD(T) quality reproduces the infrared spectrum of protonated oxalate, predicts a new weak band at 1666 cm^-1, and gives a 35.0 cm^-1 tunneling splitting for the H-transfer.","keywords":["protonated oxalate","potential energy surface","machine learning","vibrational spectroscopy","proton transfer","tunneling splitting","ring polymer instanton","VPT2"],"falsifier":"Measure the high-resolution infrared spectrum of protonated oxalate (or its deuterated analogue) and locate the split transitions between the H-transfer tunneling states. If the observed splitting is significantly different from 35.0 $cm^{-1}$ (outside combined experimental and computational uncertainty), the machine-learned PES is wrong along the transfer path. Alternatively, recompute the 1666 $cm^{-1}$ band with a PES that was not calibrated against any measured oxalate frequencies; if the band disappears, the prediction was not independent.","tokens_in":5039,"feed_emoji":"🧪","tokens_out":6291,"duration_ms":66601,"temperature":0.7,"pith_summary":"This paper aims to show that a machine-learned potential energy surface of CCSD(T) quality, tuned against new infrared measurements, captures both the vibrational framework and the proton-transfer dynamics of protonated oxalate. The key validation is that VPT2 calculations on this surface correctly identify a new low-intensity feature at 1666 $cm^{-1}$ and reproduce the measured H-transfer band around 2940 $cm^{-1}$. The paper further claims that ring-polymer instanton calculations on the same surface predict a hydrogen-transfer tunneling splitting of 35.0 $cm^{-1}$, a number that experiment can directly test. If correct, the study demonstrates that a single calibrated machine-learned surface can describe anharmonic spectroscopy, mode assignment, and quantum tunneling in a strongly hydrogen-bonded ion.","feed_headline":"Machine-learned surface pins oxalate proton tunneling at 35 cm^-1","feed_subtitle":"VPT2 on the same surface also predicts the weak 1666 cm^-1 band that experiment confirms.","key_machinery":"The central object is a machine-learned potential energy surface (a regression fit to CCSD(T) electronic energies and forces, likely a neural network) calibrated to new infrared measurements of protonated oxalate. It supplies the potential for three separate calculations: second-order vibrational perturbation theory (VPT2), which assigns the anharmonic vibrational spectrum; molecular dynamics, which characterizes the H-transfer motion; and ring polymer instanton theory with higher-order corrections, which gives the tunneling splitting from the barrier crossing path. The surface is the single entity that ties all three predictions together.","core_discovery":"The central claim is that one machine-learned potential energy surface of CCSD(T) accuracy, fit with the help of new spectroscopic results, simultaneously accounts for the observed vibrational spectrum of protonated oxalate and predicts a quantitative tunneling splitting for the intramolecular H-transfer. VPT2 on this surface assigns the framework modes and the H-transfer feature, including a newly observed low-intensity band at 1666 $cm^{-1}$ that the calculations anticipated. The broad absorption from 2600 to 3200 $cm^{-1}$ is attributed to the H-transfer motion riding on a background of combination bands, with the COH bend playing the largest role. For the deuterated isotopologue, both VPT2 and m","pith_inferences":["The tunneling splitting prediction at 35.0 cm^-1 is a benchmark unique to this PES: any competing surface for oxalate that yields a different value can be rejected by one high-resolution experiment, so the number becomes a positive test for the machine-learning training protocol.","If the 1666 cm^-1 band was not part of the calibration data, then its correct prediction signals that the model learned genuine physics rather than merely reproducing the fit set; if it was part of the fit, the paper's validation claim would weaken — a distinction the authors do not spell out.","The methodology should transfer to other ions with intramolecular hydrogen bonds, where anharmonic couplings and tunneling compete; the COH-bend's prominent role suggests that mode-specific relaxation pathways may be observable in time-resolved infrared experiments.","The manuscript text supplied with this submission is a fluid-dynamics paper on natural convection in vertical channels and does not match the abstract's topic; the extraction above therefore rests entirely on the abstract's statements."],"forward_implications":["A measured tunneling splitting in protonated oxalate near 35.0 cm^-1 would confirm the machine-learned surface's fidelity along the proton-transfer path.","The 1666 cm^-1 band, if verified independently, provides a sensitive spectral probe of the framework's anharmonicity in strongly hydrogen-bonded anions.","The assignment of the 2600–3200 cm^-1 broad feature to H-transfer plus COH-bend combination bands changes how such bands are interpreted in carboxylate and oxalate systems.","The same fitted surface, having passed these spectral tests, can be reused for dynamics and tunneling in other isotopologues with confidence."],"supporting_citations":[],"fun_headline_variants":["ML surface nails oxalate proton tunneling and IR bands","Protonated oxalate: one ML surface predicts spectra and tunneling","Machine-learned PES hits 35 cm^-1 tunneling and 1666 cm^-1 IR band","CCSD(T)-grade ML sim predicts oxalate proton dynamics","Tunneling splitting 35 cm^-1 from ML surface matches experiment"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The agreement with experiment is treated as a validation, but the PES was 'calibrated against the results of new spectroscopic measurements'; if the bands used to judge the fits—especially the new 1666 $cm^{-1}$ feature—were included in the calibration data, then the agreement is a property of the fit rather than an independent prediction, and the 35.0 $cm^{-1}$ tunneling benchmark inherits that uncertainty.","fun_headline_variants_meta":{"raw":{"variants":["ML surface nails oxalate proton tunneling and IR bands","Protonated oxalate: one ML surface predicts spectra and tunneling","Machine-learned PES hits 35 cm^-1 tunneling and 1666 cm^-1 IR band","CCSD(T)-grade ML sim predicts oxalate proton dynamics","Tunneling splitting 35 cm^-1 from ML surface matches experiment"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000683,"raw_usage":{"total_tokens":2952,"prompt_tokens":776,"completion_tokens":2176,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":2080}},"tokens_in":520,"tokens_out":2176,"duration_ms":17067,"temperature":1.0,"reasoning_tokens":2080,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T22:43:24.974514+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the high-resolution infrared spectrum of protonated oxalate (or its deuterated analogue) and locate the split transitions between the H-transfer tunneling states. If the observed splitting is significantly different from 35.0 $cm^{-1}$ (outside combined experimental and computational uncertainty), the machine-learned PES is wrong along the transfer path. Alternatively, recompute the 1666 $cm^{-1}$ band with a PES that was not calibrated against any measured oxalate frequencies; if the band disappears, the prediction was not independent.","supporting_citations":[],"review_version":1}