{"id":"bd4ea0f7-ee1f-407a-9548-5110980c12c0","arxiv_id":"2610.19846","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Hybrid reverse Monte Carlo starting from a melt-quenched glass provides amorphous structures with close fit to neutron scattering data and low energy, while preserving coordination chemistry, in contrast to pure melt-quench or pure reverse Monte Carlo.","lead":"This paper benchmarks four computational routes to build atomic models of an amorphous solid electrolyte, comparing how well each reproduces measured neutron scattering data, the predicted energy, and local chemical structure. The hybrid approach, which refines a melt-quenched glass against the scattering data using a foundation potential, emerges as the best practical choice because it simultaneously matches the measurement and preserves chemically realistic bonding.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The foundation potential (MACE-MH-1) is validated only on 55-atom disordered configurations (MAE 3.1 meV/atom), leaving its accuracy for medium-range correlations in this specific 2236-atom amorphous oxychloride composition untested, which is load-bearing because the potential generates the…","rationale":"The reader's weakest_assumption correctly identifies the foundation potential's accuracy as the load-bearing concern. The paper acknowledges the limited validation (only 55-atom test sets) and states in caveats that the potential may not capture medium-range correlations. However, the conclusion explicitly recommends the same type of potential for practical use. This makes the potential's reliability the single most important unresolved issue. The concrete test (DFT validation of relative energies and forces) would directly resolve whether the energy differences between routes are physically meaningful. I agree with the reader that the manuscript is a sound contribution—methodologically transparent, with multiple replicates and honest caveats—so the conditional acceptance is appropriate. No adjustment to the verdict is needed.","tokens_in":21406,"tokens_out":3465,"duration_ms":38038,"concrete_test":"Run r2SCAN DFT calculations (or another consistent level) on 2–3 representative configurations from each route: (1) a melt-quench glass, (2) the best hybrid structure from the MQ input (inline and alternating), (3) a pure RMC final structure. Compute total energies and compare with MACE-MH-1 predictions. Additionally, compute forces on a subset of atoms (e.g., 200 atoms) to check if the potential accurately captures local relaxations. If the relative energy ordering between routes changes by more than 0.05 eV/atom or the force MAE exceeds 50 meV/Å, then the energy-based comparisons in Fig. 3c are unreliable and the central argument weakens.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that hybrid refinement of a melt-quench glass is the best practical route—rests on the foundation potential (MACE-MH-1 with r2SCAN head) being accurate for the specific amorphous oxychloride composition Li0.9NbO0.9Cl4.1. This potential is used to (i) generate the melt-quench glass reference, (ii) score every structure in pure and hybrid refinement (energy term in inline scheme, relaxations in alternating scheme), and (iii) compare energies across routes. The paper's own validation (Section II.B and Supplementary Sec. S1) reports a mean absolute error of only 3.1 meV/atom against plane-wave r2SCAN for a set of disordered 55-atom configurations. However, 55-atom cells cannot capture medium-range order beyond ~6 Å, which is precisely the range where the hybrid refinement claims to recover packing information from the fit (Fig. 5, inset). The potential's accuracy for correlations at 6–15 Å and for the 2236-atom cell used throughout is untested. The paper acknowledges this risk (Section IV caveats, and sentence [b179]: \"an MLIP trained mostly on small periodic cells, which cover local bonding but not medium-range correlations\"), but does not provide evidence that the potential's energy ranking between different structural families (glass, pure RMC, hybrid) is physically meaningful. If the potential systematically overstabilizes or understabilizes certain medium-range arrangements—a plausible failure mode for a potential trained on small crystalline and sparse disordered snapshots—then the relative energies in Fig. 3c (the central quantitative evidence) could be misleading. For instance, the 0.1 eV/atom gap between hybrid structures and the glass might not reflect true physical preference, and the claimed superiority of the MQ-input hybrid could be an artifact of a potential that penalizes connectivity differences incorrectly.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper benchmarks four routes to an atomistic structural model of amorphous Li0.9NbO0.9Cl4.1: melt-quench (MQ) molecular dynamics, pure reverse Monte Carlo (RMC), and two hybrid RMC schemes (inline and alternating). All routes use a single neutron pair distribution function, a single 2236-atom cell, and a single foundation MLIP (MACE-MH-1 with r2SCAN head) as the energy reference. MQ alone gives the correct local chemistry but no medium-range order (R²=0.29). Pure RMC fits the data to the counting-noise ceiling (R²=0.99) but at >1.3 eV/atom above the glass, with niobium losing about half its chlorine. Both hybrid schemes, started from the MQ glass, reach R²≈0.92 within 0.1 eV/atom of the glass energy and preserve the Nb coordination chemistry. The authors conclude that hybrid refinement of a melt-quench glass is the practical route to structures that carry both local chemistry and medium-range order, and they recommend it as the starting point for ion-transport modeling.","tokens_in":21877,"tokens_out":2931,"duration_ms":33070,"significance":"If the result holds, the paper provides the first systematic, cell-size- and potential-constant comparison of MQ, pure RMC, and two hybrid schemes for an amorphous solid electrolyte. The demonstration that hybrid refinement recovers medium-range order that MQ misses while maintaining an energy penalty two orders of magnitude smaller than pure RMC is practically valuable. The finding that the input structure largely determines the polyhedral connectivity, even when the fit and energy are nearly identical, is an important caution for the field. The paper is transparent about its caveats (crystalline fraction, uneven neutron contrast, potential validation) and makes the code and data available, which supports reproducibility.","major_comments":[{"comment":"The foundation potential (MACE-MH-1 r2SCAN head) is validated only on 55-atom disordered configurations (MAE 3.1 meV/atom). The potential is used to generate the MQ glass, to score every structure in all routes, and to perform relaxations in the alternating scheme. The 55-atom cell cannot capture correlations beyond ~6 Å, which is precisely where the hybrid schemes claim to recover packing information (Fig. 5, inset). The paper acknowledges this risk (sentence b179, Discussion), but does not provide evidence that the potential's energy ranking between the MQ glass, pure RMC, and hybrid structures is physically meaningful for the 2236-atom cell. A test against DFT (r2SCAN) on a subset of configurations (e.g., one from each route, or on larger representative cells) would substantially strengthen the central claim. If such calculations are too expensive, the limitation should be stated more prominently as a caveat on the quantitative energy comparisons.","section":"Section II.B and Supplementary Sec. S1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The coordination-number (CN) constraint on Nb–O uses the distribution from the MQ glass as the target. All hybrid schemes appear to include this constraint (Methods describe it as part of the RMC loop, and it is active for pure RMC in Fig. 2 and for hybrid schemes in Figs. 3–4; please confirm explicitly). If the CN constraint is applied to the hybrid routes, then the claim that they 'keep the coordination chemistry of the glass' is partly enforced by the objective function, not solely a consequence of the energy term. The paper should state unambiguously which schemes use the CN constraint, and should report hybrid runs without it (or explain why they are omitted) to separate the effect of the constraint from that of the energy term.","section":"Section II.D.e and Section III.C","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The inline scheme's energy is still decreasing at the end of the fixed budget (b305: 'its energy still decreases slightly at the end of the budget, so the inline energies reported here are upper bounds'). This raises the question of whether the energy-R² trade-off (Fig. 3c) has converged. The authors should either extend the inline runs until the energy trace is stationary (or nearly so) or explicitly quantify the remaining drift and its likely impact on the reported energy relative to the glass.","section":"Section II.E.a and Supplementary Sec. S5","authors_acknowledge":"It costs an MLIP call for every move that passes the hard- sphere filter, and its energy still decreases slightly at the end of the budget, so the inline energies reported here are upper bounds (Supplementary Sec. S5).","authors_acknowledge_p":0.73}],"minor_comments":[{"comment":"The marker for pure RMC spans four variants at 1.3–1.5 eV/atom. It would be helpful to label which of the four (with/without constraint, from MQ/random) gives which energy, or to use separate markers.","section":"Figure 3c","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The hard-sphere table is derived from the MQ glass (0.95 times the shortest observed distance). This ties even the pure RMC runs to the glass's closest approaches. The effect of this choice on the energy and coordination of pure RMC should be mentioned in the discussion.","section":"Section II.D.c","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The green band in Fig. 2a is stated as one sample standard deviation over five glasses. However, the glasses are generated from the same melt at different seeds; the band may underestimate the true variance of the MQ ensemble. A brief comment on this would be appropriate.","section":"Section III.A, Fig. 2a","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The sentence 'The barrier a mobile ion encounters is set by the size and connectivity of the sites' cites Ref. [7]. It would be helpful to note that this bond-valence analysis is relevant for the Li positions, which the paper acknowledges are poorly constrained (b318).","section":"Section IV, last paragraph","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Several references to 'Discussion' (e.g., line after b179, b314) are forward references to the section that follows Results. This is acceptable, but the Discussion section is not labeled in the provided text; please ensure the final manuscript has clear section headings.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Dear colleague,\n\nThis paper does what it says: compare melt-quench MD, pure RMC, inline HRMC, and alternating HRMC on one oxychloride (Li0.9NbO0.9Cl4.1) with one cell, one MLIP, and one neutron PDF. The main result—that the input structure largely determines the polyhedral network even when fit and energy are nearly equal—is new and directly useful for practitioners. The benchmarking is thorough: five independent chains per condition, uncertainties reported, and the hybrid schemes are described clearly enough to reproduce.\n\nThe authors are unusually transparent about caveats. They note the sample is only ~55% amorphous, that the MLIP validation is limited to small disordered cells (MAE 3.1 meV/atom on 55 atoms), that Li contrast is poor, and that inline energies are upper bounds. This honesty strengthens the paper.\n\nThe stress-test concern about foundation-potential accuracy is real but not fatal. All structures are scored with the same potential, so relative rankings between routes are meaningful even if absolute energies carry systematic error. The authors acknowledge this risk (Section IV) and recommend user-side validation. For comparing modeling routes—the stated purpose—the potential is adequate.\n\nWeaknesses: the absolute structures (especially Li positions) are model-dependent; the study covers only one material and one measurement, so generalizability to other systems is untested. But the paper does not overclaim generality—it offers a practical workflow.\n\nWho this is for: materials modellers working on amorphous solid electrolytes. It deserves a serious referee and publication after minor revision (e.g., add a sentence on how one might extend the MLIP validation). I would cite it for the head-to-head comparison and the honest discussion.\n\nYes to peer review.","headline":"A clean, honest benchmark of amorphous structure modeling routes on an oxychloride electrolyte; the limitations are stated clearly and the central practical recommendation is well-supported.","tokens_in":22462,"tokens_out":1895,"duration_ms":19540},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Hybrid refinement of a melt-quench glass fits neutron scattering to R²=0.92 within 0.1 eV/atom while preserving local chemistry.","keywords":["amorphous solid electrolyte","reverse Monte Carlo","hybrid reverse Monte Carlo","foundation potential","MACE","neutron total scattering","pair distribution function","lithium niobium oxychloride"],"falsifier":"Compute the same hybrid refinements using a different foundation potential (e.g., the MACE-MH-1 PBE head or a system-specific MLIP trained on DFT data for amorphous oxychlorides) and check whether the energy gap between hybrid structures and the melt-quench glass changes significantly, or whether the niobium coordination can be preserved in pure RMC with a stronger set of constraints including bond-angle restraints.","tokens_in":22898,"feed_emoji":"🔬","tokens_out":4787,"duration_ms":54415,"temperature":0.7,"pith_summary":"Amorphous alkali oxychlorides are promising fast-ion conductors, but their atomic structures cannot be read directly from scattering data. The authors compare four modeling routes—melt-quench molecular dynamics, pure reverse Monte Carlo, and two hybrid reverse Monte Carlo schemes—on a single neutron pair distribution function for Li₀.₉NbO₀.₉Cl₄.₁. Melt-quench glass has the lowest energy but fits the data poorly (R²=0.29). Pure RMC fits the data extremely well (R²=0.99) but at an energy penalty of over 1.3 eV/atom and destroys the niobium coordination chemistry. Both hybrid schemes, starting from the melt-quench glass, reach R²=0.92 within 0.1 eV/atom of the glass energy while preserving the local chemistry. The paper recommends quenching a glass with a foundation potential and refining it against the measurement with a hybrid scheme, and shows that the input structure largely determines the final network connectivity.","feed_headline":"Hybrid RMC plus melt-quench glass solves amorphous structure riddle","feed_subtitle":"Combined energy and scattering refinement reaches R²=0.92 within 0.1 eV/atom of the glass while preserving coordination chemistry.","key_machinery":"The key machinery is the hybrid reverse Monte Carlo (HRMC) method, which adds a potential-energy term from the MACE-MH-1 foundation potential (r²SCAN head) to the standard RMC acceptance criterion (inline) or through periodic relaxations (alternating). The MLIP provides an energy score that prevents the structural degradation seen in pure RMC, while the fit to the scattering data supplies medium-range order missing from the melt-quench glass. The melt-quench glass, generated by the same foundation potential, provides a low-energy starting point with correct short-range bonding that the refinement preserves.","core_discovery":"The core discovery is that hybrid reverse Monte Carlo (inline or alternating) combined with a melt-quench glass input achieves a fit to the neutron pair distribution function (R²=0.92) that is far better than the glass alone (R²=0.29) and an energy (within 0.1 eV/atom of the glass) that is far lower than pure RMC (1.3–1.5 eV/atom above), while keeping the niobium coordination chemistry intact—pure RMC strips about half of the chlorine from the first shell. The two hybrid schemes differ in computational cost and final energy, but both converge to similar fit and energy from the glass input. Structures refined from a random packing of isolated units fit slightly better (R²≈0.93) but lie 0.06–0.14 eV/atom higher in energy and exhibit far fewer polyhedral connections, showing that the input determines the network topology even when the overall fit and energy are close.","pith_inferences":["The reliance on a single MLIP (MACE-MH-1 r²SCAN) means that the results should be tested with other foundation potentials (e.g., MACE-MH-1 PBE or OMat24) to confirm robustness against potential biases in the energy surface.","Because lithium positions are weakly constrained by the neutron data (only 6.3% of the neutron weight), the resulting models are not directly reliable for lithium transport paths without further validation via ab initio molecular dynamics or NMR.","The 55% amorphous fraction in the sample introduces a potential bias from the crystalline component; repeating the study on a fully amorphous sample would clarify the true medium-range order of the glassy phase and test whether the hybrid refinement recovers it completely."],"forward_implications":["This hybrid route can be applied to other amorphous solid electrolytes where total scattering data and a foundation potential are available, enabling atomistic models for ion-transport simulations.","The method separates the roles of energy and diffraction data in structure determination: the former fixes local chemistry, the latter fixes medium-range packing.","The input structure is part of the method; a melt-quench glass is recommended because it provides a physically plausible network that the refinement does not erase.","For compositions without a known crystal structure, crystal structure prediction or generative models can supply the starting point for melt-quench dynamics.","The resulting structures let future studies analyze site geometry, polyhedron connectivity, and free volume to understand ionic transport barriers."],"supporting_citations":[{"why":"Introduces the reverse Monte Carlo (RMC) method, which is the base refinement technique used throughout the paper.","marker":"[20]"},{"why":"Describes the hybrid reverse Monte Carlo (HRMC) scheme where energy is added to the acceptance rule, adopted here as the inline scheme.","marker":"[25]"},{"why":"Presents the MACE foundation potential (MACE-MH-1), which serves as the energy reference for all routes in this work.","marker":"[17]"},{"why":"Provides the MatPES r²SCAN dataset and head used to validate the MLIP against plane-wave DFT for disordered configurations.","marker":"[15]"},{"why":"Reports the crystal structure of LiNbOCl₄, used to construct the starting supercell for melt-quench molecular dynamics.","marker":"[8]"},{"why":"Describes the mechanochemical synthesis and neutron total-scattering measurement of Li₀.₉NbO₀.₉Cl₄.₁ that serves as the experimental reference.","marker":"[19]"},{"why":"Introduces force-enhanced atomic refinement, the basis of the alternating hybrid scheme in this paper.","marker":"[29]"},{"why":"Provides the bond-valence rationale for relating polyhedron connectivity to ion transport, motivating the structural analysis performed here.","marker":"[7]"}],"fun_headline_variants":["Hybrid RMC melts away amorphous structure ambiguity","Melt-quench glass plus RMC hits R²=0.92 at glass energy","Pure RMC strips chlorine; hybrid keeps coordination intact","Two hybrid routes, same fit, different networks—input matters"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The entire comparison depends on the accuracy of the MACE-MH-1 foundation potential (r²SCAN head) for this disordered oxychloride composition: if the MLIP gives unreliable energies or forces for this system, the energy comparisons and the melt-quench glass itself may be misleading.","fun_headline_variants_meta":{"raw":{"variants":["Hybrid RMC melts away amorphous structure ambiguity","Melt-quench glass plus RMC hits R²=0.92 at glass energy","Pure RMC strips chlorine; hybrid keeps coordination intact","Two hybrid routes, same fit, different networks—input matters"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":7.6e-05,"raw_usage":{"cost":7.635306e-05,"is_byok":false,"cost_details":{"upstream_inference_cost":7.635306e-05,"upstream_inference_prompt_cost":5.064738e-05,"upstream_inference_completions_cost":2.570568e-05},"total_tokens":1413,"prompt_tokens":1127,"completion_tokens":286,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":213}},"tokens_in":1127,"tokens_out":286,"cost_source":"openrouter_usage","duration_ms":4277,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":213,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-15T13:39:07.956531+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Compute the same hybrid refinements using a different foundation potential (e.g., the MACE-MH-1 PBE head or a system-specific MLIP trained on DFT data for amorphous oxychlorides) and check whether the energy gap between hybrid structures and the melt-quench glass changes significantly, or whether the niobium coordination can be preserved in pure RMC with a stronger set of constraints including bond-angle restraints.","supporting_citations":[{"cited_title":"Reverse Monte Carlo simulation: A new technique for the determination of dis- ordered structures, Molecular Simulation1, 359","cited_arxiv_id":null,"evidence_quote":"Introduces the reverse Monte Carlo (RMC) method, which is the base refinement technique used throughout the paper."},{"cited_title":"Hybrid approach for generating realistic amorphous carbon structure using Metropolis and reverse Monte Carlo, Molecular Simula- tion28, 927","cited_arxiv_id":null,"evidence_quote":"Describes the hybrid reverse Monte Carlo (HRMC) scheme where energy is added to the acceptance rule, adopted here as the inline scheme."},{"cited_title":"P.; Simm, G.; Ortner, C.; Cs´ anyi, G","cited_arxiv_id":null,"evidence_quote":"Presents the MACE foundation potential (MACE-MH-1), which serves as the energy reference for all routes in this work."},{"cited_title":"arXiv preprint arXiv:2503.04070 , year=","cited_arxiv_id":null,"evidence_quote":"Provides the MatPES r²SCAN dataset and head used to validate the MLIP against plane-wave DFT for disordered configurations."},{"cited_title":"Critical Role of Framework Flex- ibility and Disorder in Driving High Ionic Conductivity in LiNbOCl4, Journal of the American Chemical Society 146, 17158","cited_arxiv_id":null,"evidence_quote":"Reports the crystal structure of LiNbOCl₄, used to construct the starting supercell for melt-quench molecular dynamics."},{"cited_title":"J.-c. Tseng, X. Zhang, P. Yu, W. Tang, J. Lei, S. Zhang, P. Wang, Y. Li, M. Liu, W. Yin, L. Wang, S. Han, W. Xia, Y. Zhao, and J. Zhu, Rapid mechanochemical synthesis of oxyhalide superionic con- ductor: Time-resolved structural evolution, Small Meth- ods10, e00947","cited_arxiv_id":null,"evidence_quote":"Describes the mechanochemical synthesis and neutron total-scattering measurement of Li₀.₉NbO₀.₉Cl₄.₁ that serves as the experimental reference."},{"cited_title":"Realistic inversion of diffraction data for an amorphous solid: The case of amorphous silicon, Physical Review B 94, 235208","cited_arxiv_id":null,"evidence_quote":"Introduces force-enhanced atomic refinement, the basis of the alternating hybrid scheme in this paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the bond-valence rationale for relating polyhedron connectivity to ion transport, motivating the structural analysis performed here."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}