{"id":"27e612c9-9682-4713-9d69-cfb79ed4deb4","arxiv_id":"2508.01020","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":2,"one_line_summary":"A user-friendly code that reconstructs potential energy surfaces of crystals via high-order Taylor expansion and machine-learning-based extraction of interatomic force constants.","lead":"A new open-source code, Pheasy, extracts interatomic force constants of arbitrary high order from force-displacement data using machine learning. It aims to push phonon calculations beyond the usual third-order anharmonicity and connect diverse simulation platforms.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The abstract promises arbitrarily high-order Taylor reconstruction and optimal IFC extraction, but gives no identifiability or regularization evidence; without it, high-order IFCs may be order-dependent and nonunique.","rationale":"The abstract-only review left the paper UNVERDICTED, and the stress-test pass has no more information than the abstract. I read the strongest claim as a promise of accurate, effectively order-independent IFCs from finite force-displacement data. The single load-bearing condition is that the inverse problem is well-posed: the Taylor coefficients are identifiable from the sampled dataset and stable under truncation. The abstract itself flags the combinatorial explosion of higher-order IFCs, so this is not a manufactured tension. Without a stated regularization or selection scheme, an ML fit of a high-order polynomial to finite data can be ill-conditioned and nonunique; 'arbitrarily high order' then becomes a property of the implemented ansatz, not of the physical reconstruction. Since no methods, equations, or repository are available, I cannot verify whether Pheasy addresses this with symmetry-adapted bases, sparsity, or validation-based truncation. The concrete test would settle it by checking order-convergence and held-out force errors on one benchmark material. As no defect is demonstrated, the existing UNVERDICTED verdict is unchanged; if the test reveals order dependence, the verdict should move toward CONDITIONAL or REJECT.","tokens_in":716,"tokens_out":3082,"duration_ms":40530,"concrete_test":"Download Pheasy and its benchmark data; for one prototypical material (e.g., silicon or MgO), fit IFCs at maximum Taylor orders 3, 4, 5, and 6 from the same fixed DFT force-displacement set. Report (i) whether low-order IFCs converge with increasing maximum order, (ii) the condition number or singular spectrum of the fitting design matrix, and (iii) force errors on held-out displacement configurations excluded from fitting. If low-order IFCs shift materially with truncation order, or held-out errors do not improve, then the arbitrary-order and optimal-extraction claims are not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that Pheasy reconstructs the potential energy surface via a Taylor expansion of arbitrarily high order and that benchmarks identify the optimal IFC-extraction approach. The abstract itself notes the combinatorial explosion in the number of higher-order IFCs, yet it does not state how a finite force-displacement dataset uniquely determines those coefficients. In high-order polynomial fitting to finite, symmetry-reduced data, the least-squares or ML design matrix is typically ill-conditioned or rank-deficient; without explicit regularization, basis selection, or a physical prior, the estimator returns one of many interpolants. If such constraints are absent, fitted IFCs will depend on the maximum Taylor order and on the training set, which would contradict the claim of accurate PES reconstruction and undermine the asserted identification of an optimal extraction scheme. This concern is load-bearing because the abstract's strongest claim rests on the reliability of extracted IFCs. Since the full text is unavailable, this is a pointed question rather than a demonstrated error; the absence of stated evidence is the issue, not the character of the work.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents Pheasy, a first-principles program for lattice dynamics that reconstructs the potential energy surface of crystalline solids via a Taylor expansion of arbitrarily high order, extracts interatomic force constants (IFCs) from force-displacement data using machine-learning algorithms, and computes harmonic and anharmonic phonon properties. The abstract reports successful application to three prototypical examples and claims that the benchmarks identify the optimal IFC-extraction approach and provide general guidelines for high-fidelity lattice-dynamical simulations. The full text was not available for this review; the assessment is based solely on the abstract.","tokens_in":923,"tokens_out":3588,"duration_ms":39537,"significance":"If Pheasy delivers what the abstract promises, it would be a valuable community resource: high-order IFCs are central to anharmonic lattice dynamics, and a user-friendly, platform-connecting code with validated extraction tools could reduce the uncertainty noted in existing schemes. The claimed 'optimal approach' for IFC extraction, if substantiated by quantitative comparisons, would be a useful practical contribution. However, the abstract alone contains no numerical results, error metrics, or methodological details, so the significance is currently prospective rather than established. The paper does not appear to include machine-checked proofs, but reproducibility would depend on code/data release, which the abstract does not mention.","major_comments":[{"comment":"The abstract claims accurate reconstruction of the potential energy surface via a Taylor expansion of 'arbitrarily high order' and efficient extraction of IFCs, but it does not state how a finite force-displacement dataset uniquely determines high-order IFCs, which are subject to combinatorial explosion. In high-order polynomial fitting, the design matrix is typically rank-deficient or ill-conditioned, causing the extracted IFCs to depend on the truncation order and on the regularization or basis-selection scheme. The manuscript must specify the identifiability conditions, regularization strategy, and validation against independent force data to support this central claim.","section":"Abstract (first two sentences)"},{"comment":"The assertion that the benchmarks 'identified the optimal approach for IFC extractions' is not supported by any quantitative comparison in the abstract. The paper should report convergence of IFCs with respect to displacement-set size, Taylor order, and hyperparameter choices, and define the optimality metric (e.g., RMS force error, phonon dispersion error, or thermal conductivity error) with respect to a reference.","section":"Abstract (benchmark sentence)"},{"comment":"The abstract calls the calculations 'parameter-free' while simultaneously invoking 'advanced machine-learning algorithms'; if these algorithms involve hyperparameters, the phrase is misleading. Please clarify what 'parameter-free' means here, and state which hyperparameters are user-set and how they are selected (e.g., cross-validation), or remove the term from the abstract.","section":"Abstract (opening phrase)"}],"minor_comments":[{"comment":"The program name 'Pheasy' is not explained; a sentence on the intended meaning or acronym would help readability.","section":"Abstract"},{"comment":"The phrase 'advanced machine-learning algorithms' is vague; naming the specific algorithms (e.g., linear regression with L1/L2 regularization, neural networks, Gaussian processes) would improve the abstract and allow experts to judge the approach.","section":"Abstract"},{"comment":"The abstract mentions 'broad research community' and connecting 'diverse phonon simulation platforms' but does not state the license, repository access, or interfaces; a sentence on availability would be useful.","section":"Abstract"}],"recommendation":"uncertain","confidential_remarks":"The review is based solely on the abstract because the full text was not provided; this necessarily limits the confidence of the assessment. I recommend obtaining the full manuscript before making an editorial decision, or, if the journal permits, treating this report as provisional."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a software paper promising arbitrary-order interatomic force constant (IFC) extraction with machine learning, which is a genuine need in the phonon community. The abstract reads coherently and the authors are credible (Marzari's group). What's new is the claim of going beyond the usual third-order limit in a general-purpose, user-friendly code, plus benchmarks that apparently identify the best extraction scheme. If the code ships with the paper, that is real, reproducible evidence and should be treated as such by referees.\n\nI can't verify the strong claims because we only have the abstract. That's the main soft spot: no quantitative benchmarks, no error metrics, no methodological detail. The stress-test concern about identifiability is well aimed. High-order polynomial fitting to finite force-displacement data can be ill-posed; without regularization, basis selection, or a physical prior, the fitted IFCs may depend on the maximum Taylor order and the training set. The abstract acknowledges the combinatorial explosion but doesn't say how the code keeps the problem well-conditioned. That is a pointed question, not a demonstrated error. The authors may well use sparsity-promoting ML methods or symmetry-adapted bases, which would address it. But the abstract gives no hint, so the central reliability claim is unverified.\n\nWorth noting what the paper does well even from the abstract: it offers a unified platform, attempts to standardize IFC extraction, and gives practical guidelines. Those are useful contributions for the community, provided the underlying validation is solid.\n\nMy recommendation: send it to peer review, but the editor should require the full code, the input/output data for the three benchmark examples, and a direct comparison with existing codes like phono3py and hiphive. The referees should specifically probe the identifiability and convergence issues. If the code is open-source and the benchmarks reproduce, this could be a genuinely useful tool. If the code isn't made available or the benchmarks are cherry-picked, the paper falls flat. I'd want to see the full text before citing it, but it's worth a serious referee.","headline":"A plausible and potentially useful phonon-code paper, but the abstract alone can't support the accuracy and optimality claims; the identifiability question is real and must be answered in review.","tokens_in":1405,"tokens_out":1126,"would_cite":false,"duration_ms":17654,"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":"Pheasy reconstructs crystal potential energy surfaces at arbitrarily high Taylor order, making high-order anharmonic phonon calculations practical.","keywords":["phonons","interatomic force constants","lattice dynamics","anharmonicity","machine learning","potential energy surface","thermal transport","first-principles calculations"],"falsifier":"Take a strongly anharmonic crystal such as PbTe, compute its lattice thermal conductivity with Pheasy at successive Taylor orders up to the practical maximum, and compare with experimental values and direct molecular dynamics; if the predictions fail to converge or disagree systematically, the arbitrary-order reconstruction is not valid for that material.","tokens_in":554,"feed_emoji":"⚛️","tokens_out":5843,"duration_ms":50786,"temperature":0.7,"pith_summary":"Pheasy aims to make parameter-free lattice-dynamics calculations tractable beyond third-order anharmonicity, where the number of interatomic force constants explodes combinatorially. The paper claims that the code reconstructs the potential energy surface of a crystal through a Taylor expansion of arbitrarily high order, extracts the force constants from force–displacement data using machine-learning algorithms, and computes harmonic and anharmonic phonon properties. Three benchmark systems show the extracted force constants can be used for anharmonic lattice dynamics and thermal transport, and the benchmarks identify the most robust extraction scheme. If correct, the approach would extend first-principles phonon calculations to strongly anharmonic materials and provide a community platform for phonon simulations.","feed_headline":"Pheasy code captures phonons at any Taylor order","feed_subtitle":"High-order interatomic force constants from machine learning make strongly anharmonic lattice dynamics practical.","key_machinery":"The central object is the Taylor expansion of the Born–Oppenheimer potential energy surface $V(\\mathbf{R})$ in atomic displacements around equilibrium, $V = \\sum_{n\\ge 0} \\frac{1}{n!}\\Phi^{(n)} \\cdot (\\Delta \\mathbf{R})^n$, whose coefficients $\\Phi^{(n)}$ are the interatomic force constants. Pheasy makes this expansion usable at high order by exploiting lattice symmetry to reduce the number of independent coefficients, sampling force–displacement data from first-principles calculations, and fitting the coefficients with machine-learning algorithms that remain stable as the order grows. The resulting force constants then feed calculations of vibrational spectra and thermal transport.","core_discovery":"The paper presents Pheasy, a program that reconstructs the potential energy surface of a crystalline solid as a Taylor expansion in atomic displacements of arbitrarily high order, with the interatomic force constants as expansion coefficients. From force–displacement datasets, typically from density-functional theory, the code extracts the coefficients using machine-learning algorithms and then computes phonon-related properties. Benchmarks on three prototypical materials demonstrate that the resulting force constants can describe anharmonic lattice dynamics and thermal transport, and identify the most reliable extraction scheme among existing approaches.","pith_inferences":["If the arbitrary-order Taylor reconstruction is as reliable as the benchmarks suggest, the same fitting machinery could be applied to machine-learned interatomic potentials directly, removing the need to fit each force-constant order separately.","The convergence of the Taylor expansion likely depends on the material and the sampled displacement range, so an automatic convergence check with respect to expansion order would strengthen the claim of parameter-free accuracy.","For defective or disordered crystals, the combinatorial growth of force constants may re-emerge, and extensions exploiting local symmetry or low-rank structure could be needed."],"forward_implications":["High-order force constants become computationally accessible, so anharmonic effects beyond cubic terms can be included in parameter-free lattice-dynamics calculations.","Reliable extraction of higher-order force constants improves predictions of thermal transport, phase transitions, and thermodynamic properties of strongly anharmonic materials.","The identified optimal extraction scheme provides a benchmark for other phonon codes, reducing the scatter seen across existing force-constant fitting approaches.","A modular phonon ecosystem that connects different simulation platforms lowers the barrier for non-specialists to run high-fidelity lattice-dynamics simulations."],"supporting_citations":[],"fun_headline_variants":["Pheasy: phonon physics at any anharmonic order","Machine learning extracts force constants of all orders","Pheasy: arbitrary-order Taylor terms for phonons","High-order force constants from first principles using Pheasy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Taylor expansion of the potential energy surface converges over the sampled displacement range and that the force–displacement data determine the high-order coefficients uniquely; in practice, truncating at a finite order can change the extracted force constants.","fun_headline_variants_meta":{"raw":{"variants":["Pheasy: phonon physics at any anharmonic order","Machine learning extracts force constants of all orders","Pheasy: arbitrary-order Taylor terms for phonons","High-order force constants from first principles using Pheasy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000208,"raw_usage":{"total_tokens":1365,"prompt_tokens":871,"completion_tokens":494,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":487,"completion_tokens_details":{"reasoning_tokens":430}},"tokens_in":487,"tokens_out":494,"duration_ms":6768,"temperature":1.0,"reasoning_tokens":430,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:51:59.885615+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a strongly anharmonic crystal such as PbTe, compute its lattice thermal conductivity with Pheasy at successive Taylor orders up to the practical maximum, and compare with experimental values and direct molecular dynamics; if the predictions fail to converge or disagree systematically, the arbitrary-order reconstruction is not valid for that material.","supporting_citations":[],"review_version":1}