{"id":"09da8db6-2208-46c6-b707-df7d4c9597b1","arxiv_id":"2606.20500","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Machine-learning MD produces an ideal fourfold a-Si network whose hybrid-DFT bandgap matches experiment.","lead":"This paper presents a defect-free model of amorphous silicon generated from machine-learning molecular dynamics simulations that, when computed with hybrid density functional theory, reproduces the experimental electronic bandgap. It supplies a clean atomic structure for future studies of band tails, optics, and transport in a-Si without defect complications.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"ML potential used for structure generation may embed biases that make the subsequent hybrid-DFT bandgap match non-independent of the generation protocol.","rationale":"The reader's weakest assumption directly identifies the same load-bearing point; no stronger internal inconsistency is visible from the supplied abstract.","tokens_in":1629,"tokens_out":323,"duration_ms":23336,"concrete_test":"Take the published ML-MD configuration, generate a second 512-atom cell with the WWW algorithm under identical density and quench protocol, relax both cells to the same force tolerance with the hybrid functional, and compare the resulting Kohn-Sham gaps and tail-state densities; a difference larger than the experimental uncertainty (~0.1 eV) would indicate that the reported match is protocol-dependent.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the hybrid-DFT bandgap computed on the ML-MD structure is an unbiased property of an ideal continuous-random-network a-Si. Because the atomic coordinates originate from an ML interatomic potential (Rosset et al., Nat. Commun. 2025) whose training data and loss function are not described here, any structural correlations that the potential learned (e.g., preferred ring statistics, bond-angle distributions, or medium-range order) can propagate into the electronic density of states even when the Hamiltonian is switched to hybrid DFT. The abstract provides no evidence that an identically sized cell generated by an orthogonal method (WWW, ab-initio MD, or a different ML potential) yields a statistically indistinguishable gap once evaluated at the same hybrid-DFT level.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript claims that a defect-free ('ideal') model of amorphous silicon generated via machine-learning-driven molecular-dynamics simulations (taken from Rosset et al., Nat. Commun. 2025) reproduces the experimentally observed electronic bandgap when its electronic structure is recomputed at the hybrid-DFT level. The work compares this model to the Wooten-Winer-Weaire (WWW) bond-switching model and other recent approximants, positioning the structure as a platform for studies of band tails, optical properties, and transport.","tokens_in":1808,"tokens_out":572,"duration_ms":19480,"significance":"If the central claim holds after addressing independence concerns, the result would supply a high-quality, large-scale continuous-random-network model of a-Si with a pristine gap, which is valuable for the field because traditional models often contain defects that complicate electronic-property calculations. The ML-MD route enables access to defect-free configurations at scales difficult for direct ab initio methods, and the hybrid-DFT evaluation step is a standard and appropriate choice for gap accuracy.","major_comments":[{"comment":"The structure originates from the ML interatomic potential of Rosset et al. (Nat. Commun. 2025). No test is reported in which an identically sized cell generated by an orthogonal protocol (WWW, ab-initio MD, or a different ML potential) is relaxed and evaluated at the same hybrid-DFT level to demonstrate that the reported bandgap match is independent of structural correlations learned during potential training (e.g., ring statistics or bond-angle distributions).","section":"Methods / Results (structure generation and electronic evaluation)"},{"comment":"The abstract states that the hybrid-DFT bandgap 'accurately reproduce[s] the experimentally observed electronic bandgap,' yet the provided text supplies neither the numerical gap value, its uncertainty, nor a direct comparison plot or table against experiment and the WWW reference. Without these quantities the quantitative claim cannot be assessed.","section":"Abstract and Results section"}],"minor_comments":[{"comment":"Notation for the hybrid functional (exact exchange fraction, range-separation parameters) and k-point sampling used in the DFT calculations should be stated explicitly in the methods.","section":"Methods"},{"comment":"Figure captions should include the system size (number of atoms) and the precise definition of 'defect-free' (e.g., coordination cutoff) for each model shown.","section":"Figures"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is built directly on a 2025 Nat. Commun. paper by overlapping authors; the cover letter should confirm that the electronic-structure calculation constitutes a genuinely new contribution rather than a re-analysis of already-published coordinates."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive and detailed report. The comments identify key areas where additional comparisons and quantitative clarity would strengthen the manuscript. We address each major comment below and will revise the manuscript accordingly.","responses":[{"response":"We agree that an explicit test with an orthogonal generation protocol evaluated at the identical hybrid-DFT level would help demonstrate that the bandgap result is not an artifact of structural features learned by the ML potential. In the revised manuscript we will add such a comparison: a WWW-generated cell of comparable size will be relaxed consistently with the ML potential and then evaluated at hybrid-DFT, with the resulting bandgap reported alongside the ML-MD result and experiment.","revision_made":"yes","referee_comment":"[Methods / Results (structure generation and electronic evaluation)] The structure originates from the ML interatomic potential of Rosset et al. (Nat. Commun. 2025). No test is reported in which an identically sized cell generated by an orthogonal protocol (WWW, ab-initio MD, or a different ML potential) is relaxed and evaluated at the same hybrid-DFT level to demonstrate that the reported bandgap match is independent of structural correlations learned during potential training (e.g., ring statistics or bond-angle distributions)."},{"response":"We acknowledge the omission. The revised manuscript will update the abstract to state the numerical hybrid-DFT bandgap value (with uncertainty) and its relation to experiment. We will also add a dedicated table or figure in the Results section that directly compares the gap values obtained for the ML-MD model, the WWW reference, and the experimental value.","revision_made":"yes","referee_comment":"[Abstract and Results section] The abstract states that the hybrid-DFT bandgap 'accurately reproduce[s] the experimentally observed electronic bandgap,' yet the provided text supplies neither the numerical gap value, its uncertainty, nor a direct comparison plot or table against experiment and the WWW reference. Without these quantities the quantitative claim cannot be assessed."}],"tokens_in":1314,"tokens_out":431,"duration_ms":30292,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core claim is that an ML-generated defect-free a-Si cell, when checked at hybrid DFT level, reproduces the experimental gap and offers a cleaner starting point than older models for band-tail work. That specific validation step on this structure is new relative to their 2025 Nat. Commun. paper.\n\nWhat stands out is the direct comparison to a WWW cell and the positioning of the model as a platform for transport and optical calculations. If the numbers hold up in the full text, it could be useful for people who need a large, fourfold-coordinated cell without obvious defects.\n\nThe soft spots are the missing quantitative evidence in the abstract—no reported gap value, no error bar, no plot against experiment or other models. That makes it impossible to judge how close the match actually is. More importantly, the structure originates from their own ML potential; any ring statistics or medium-range order baked into the training data could carry over into the hybrid-DFT density of states, so the agreement may not be an independent test. The stress-test note on this point is fair until the paper shows an orthogonal generation method yielding the same gap.\n\nThis is aimed at the disordered-semiconductor modeling community rather than a broad audience. Readers who already work with a-Si cells and want a defect-free reference will get the most out of it. The work is coherent on its own terms and the claim is falsifiable, so it deserves a serious referee even if revisions are needed on the quantitative comparisons and bias checks.","headline":"They take their prior ML-MD a-Si structure, run hybrid DFT on it, and claim the bandgap matches experiment while comparing to WWW models, but the abstract gives no numbers and the independence looks shaky.","tokens_in":2311,"tokens_out":390,"would_cite":false,"duration_ms":12634,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A defect-free model of amorphous silicon generated by machine-learning molecular dynamics and checked with hybrid DFT reproduces the experimental electronic bandgap.","keywords":["amorphous silicon","defect-free model","electronic bandgap","machine-learning molecular dynamics","hybrid density functional theory","continuous random network","band tails"],"falsifier":"A direct hybrid-DFT calculation on the ML-generated structure that yields a bandgap differing substantially from the accepted experimental value for a-Si would falsify the central claim.","tokens_in":2558,"feed_emoji":"","tokens_out":696,"duration_ms":15869,"temperature":0.7,"pith_summary":"The paper establishes that an 'ideal' continuous random network of a-Si, free of coordination defects, can be produced through machine-learning-driven molecular dynamics and then evaluated at the hybrid density-functional level to yield an electronic bandgap matching experiment. A sympathetic reader would care because previous models of a-Si have struggled to combine structural ideality with accurate electronic properties without introducing artificial defects or relying on empirical adjustments. If correct, the approach supplies a clean platform for examining band tails, optical absorption, and charge transport in the material without the confounding effects of dangling bonds or wrong bonds. The central object carrying the argument is the ML-generated atomic configuration whose electronic structure is recomputed independently at higher theory level.","feed_headline":"Defect-free a-Si model matches experimental bandgap","feed_subtitle":"Machine-learning MD generates an ideal continuous random network whose hybrid-DFT bandgap aligns with measured values, enabling cleaner stud","key_machinery":"The machine-learning interatomic potential that generates the defect-free a-Si configuration during molecular dynamics, followed by independent hybrid-DFT recomputation of its electronic structure.","core_discovery":"A defect-free ('ideal') model of a-Si from machine-learning-driven molecular-dynamics simulations, subsequently evaluated with hybrid-level density-functional theory computations, can accurately reproduce the experimentally observed electronic bandgap. The authors compare this model with one from the Wooten-Winer-Weaire bond-switching method and with other recent approximants to ideal a-Si, and they position the new structure as a platform for studies of band tails, optical properties, and transport.","pith_inferences":["If the ML potential is unbiased for electronic properties, the same structures could be used to benchmark cheaper semi-local functionals against hybrid results.","The approach suggests that structural ideality and correct bandgap can be achieved simultaneously, which may reduce the need for post-hoc defect insertion in device modeling.","Extension to larger cells would allow direct simulation of localized tail states and their participation in transport."],"forward_implications":["The model supplies a reference structure free of coordination defects for investigating band-tail states.","Optical and transport calculations can be performed on the same defect-free network without artificial gap states from dangling bonds.","Direct comparison with the WWW bond-switching model isolates the effect of generation method on electronic properties.","The workflow can be repeated for other amorphous semiconductors to test transferability of the ML-plus-hybrid-DFT protocol."],"fun_headline_variants":["ML simulations generate defect-free a-Si with experimental bandgap","Defect-free a-Si model from ML MD matches measured bandgap","Ideal a-Si structure reproduces experimental bandgap via hybrid DFT","New defect-free a-Si model aligns with measured bandgap"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The machine-learning potential used to create the atomic arrangement produces a configuration whose electronic bandgap, when recomputed with hybrid DFT, is not biased by the potential's training data or simulation choices.","fun_headline_variants_meta":{"raw":{"variants":["ML simulations generate defect-free a-Si with experimental bandgap","Defect-free a-Si model from ML MD matches measured bandgap","Ideal a-Si structure reproduces experimental bandgap via hybrid DFT","New defect-free a-Si model aligns with measured bandgap"]},"model":"grok-4.3","cost_usd":0.008337,"raw_usage":{"total_tokens":3736,"prompt_tokens":587,"num_sources_used":0,"completion_tokens":63,"cost_in_usd_ticks":83374500,"prompt_tokens_details":{"text_tokens":587,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3086,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":587,"tokens_out":63,"duration_ms":25547,"temperature":1.0,"reasoning_tokens":3086,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-26T16:04:36.770304+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A direct hybrid-DFT calculation on the ML-generated structure that yields a bandgap differing substantially from the accepted experimental value for a-Si would falsify the central claim.","supporting_citations":[],"review_version":1}