{"id":"bf9b5f52-0af5-42da-b0d7-7f98e079b02f","arxiv_id":"2607.22316","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Zero-shot universal MLIP molecular dynamics reproduces the heating/cooling CDW phase sequence of monolayer 1T-TaS2, including Star-of-David order loss, thermal hysteresis, and persistent α/β chiral multi-domain freezing.","lead":"Using a universal machine-learning interatomic potential with no material-specific fine-tuning, the authors simulate a 2,028-atom monolayer of 1T-TaS2 and observe the expected charge-density-wave melting sequence, thermal hysteresis, and freezing into mixed-chirality domains. The work demonstrates a scalable route to finite-temperature CDW simulations that were previously impractical with first-principles molecular dynamics.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central transition sequence rests on UMA-s-1p1's accuracy in unbenchmarked NCCDW/domain-wall configurations, while the stated energy error (0.048 eV/f.u.) exceeds the NM–CCDW enthalpy difference (0.027 eV/f.u.).","rationale":"I read the paper as a coherent computational demonstration whose central claim is conditional on the surrogate potential accurately representing the Born–Oppenheimer surface in the configurations that actually occur during the transition. The reader's weakest_assumption names exactly this: the benchmark samples only near-equilibrium displacements of the two endpoint phases, while the NCCDW coexistence regions, α/β domain walls, and boundary regions shown in Fig. 4 and Fig. S7–S8 are never benchmarked against DFT. The energy-scale mismatch (MAE 0.048 eV/f.u. vs ΔH 0.027 eV/f.u.) makes this a genuine risk rather than a stylistic quibble. The paper has real supporting evidence: deposited scripts/videos, TDEP mode stabilization consistent with the structural order parameter, and explicit statement of limitations. However, those do not close the validation gap, because the TDEP result is derived from the same MD trajectories and inherits any MLIP bias. The proposed check—DFT single-point energies on small supercells extracted from the actual MD snapshots, plus a direct domain-wall energy comparison—would settle whether the qualitative claims are robust. Since the reader already marked the paper CONDITIONAL on essentially this assumption, I would keep that verdict; my stress test does not move it. No ad hominem is intended; the critique targets the surrogate-validity premise, not the authors' conduct.","tokens_in":11147,"tokens_out":6195,"duration_ms":56617,"concrete_test":"Use the deposited MD trajectories to extract 20–30 snapshots at 300 K (NCCDW coexistence), 160 K (multi-domain), and 60 K (CCDW). From each snapshot, carve representative 156-atom 2√13×2√13 supercells containing the local motifs (SoD interior, α/β domain wall, compressed interdomain bonds). Recompute energies with PBE using the same PAW settings, cutoff, and k-mesh as Section II and compare with UMA-s-1p1 on identical structures. Report MAE per f.u. and signed errors. Separately, compute the DFT energy cost of a single α/β domain wall (cell with wall minus perfect CCDW cell) and compare to the UMA-predicted wall energy. If either error exceeds the 0.027 eV/f.u. phase-stability scale (or the DFT wall energy), the qualitative transition sequence is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim is that large-scale MD with a zero-shot universal MLIP reproduces the experimentally observed CDW transition sequence, including NCCDW-like coexistence, hysteresis, and persistent α/β multi-domains. For this to hold, UMA-s-1p1 must be accurate on exactly the configurations that define those phases: partially formed SoD clusters, α/β domain walls, compressed interdomain bonds (~3.2 Å), and transient ICDW-like clusters. The benchmark in Section III/Fig. 2 only probes random displacements of 0.01–0.15 Å from the relaxed NM and CCDW references. The reported MAE, 0.048 eV/f.u., is larger than the NM–CCDW enthalpy difference (ΔH = 0.0272 eV/f.u.) that sets the phase-stability scale. Reproducing ΔH for the two relaxed endpoints does not validate the energy surface for the intermediate and boundary structures that carry the qualitative conclusions (Fig. 4, Fig. S7, Fig. S8). If the surrogate's errors on these unseen environments are comparable to the relevant domain-wall or coexistence energies, the 290 K melting, ~30 K hysteresis, and persistent multi-domain freezing could be artifacts of the MLIP rather than the physical potential. This is a validation gap, not an internal inconsistency; the paper's own limitation statement about quantitative transition temperatures does not cover the qualitative domain-multiplication claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper benchmarks two universal machine-learning interatomic potentials (MACE-MP-0 and UMA-s-1p1) against PBE and PBE+U displacement energies for monolayer 1T-TaS2, selects UMA-s-1p1, and then uses it for 2028-atom molecular dynamics simulations. On heating from the commensurate Star-of-David (SoD) phase, the authors observe a sharp loss of SoD order near 290 K, with an intermediate nearly commensurate CDW-like state, followed by an incommensurate-like regime of transient clusters and finally the undistorted hexagonal phase near 500 K. On cooling, the transition occurs near 260 K, giving a ~30 K hysteresis; the system freezes into a multi-domain state with α and β chiral domains. Temperature-dependent effective potential (TDEP) phonon calculations from the MD trajectories show progressive stabilization of soft modes with temperature. The central claim is that a zero-shot universal MLIP, benchmarked only against DFT displacement energies, reproduces the experimentally observed CDW phase transition sequence, including the intermediate coexistence, hysteresis, and multi-domain freezing.","tokens_in":11351,"tokens_out":3164,"duration_ms":30483,"significance":"If correct, this is a valuable demonstration that universal MLIPs can capture finite-temperature lattice transformations in a correlated CDW material at a scale inaccessible to direct DFT-MD. The paper's strengths include explicit DFT benchmarking, a quantitative structural order parameter for SoD order, 2028-atom MD simulations, TDEP phonon analysis, and public availability of analysis scripts via Figshare. The zero-shot aspect is notable and would extend recent system-specific MLIP studies of NbSe2. However, the central claim depends on the surrogate's accuracy in configurations that are not benchmarked, and the reported energy error is comparable to the phase-stability energy scale. The qualitative domain-multiplication and hysteresis conclusions require additional validation or a more cautious framing.","major_comments":[{"comment":"The benchmark that justifies using UMA-s-1p1 samples only random displacements of 0.01–0.15 Å from the relaxed NM and CCDW reference structures. The mean absolute error reported for UMA-s-1p1, 0.048 eV/f.u., is larger than the NM–CCDW enthalpy difference reported in the same section, ΔH = 0.0272 eV/f.u., which sets the phase-stability scale. The intermediate NCCDW-like states, α/β domain walls, interdomain regions with compressed ~3.2 Å bonds, and transient ICDW-like clusters shown in Fig. 4 and Figs. S7–S8 are never benchmarked against DFT. If the MLIP error in those unbenchmarked environments is comparable to the domain-wall or coexistence energies, the 290 K melting, ~30 K hysteresis, and persistent multi-domain state could be artifacts of the surrogate rather than physical. The paper's limitation statement about quantitative transition temperatures does not cover these qualitative co","section":"Section III, Fig. 2; Section II.B"},{"comment":"The SoD order parameter is purely geometric: it classifies Ta atoms based on Ta–Ta distances and coordination shells. The paper repeatedly refers to 'CDW' phases, but no electronic charge modulation is measured in the MD simulations. The geometric SoD pattern may not uniquely correspond to the electronic CDW order, especially at finite temperature and at domain boundaries where the classification uses an adaptive tolerance and accepts partially formed patterns. To support the claim that the simulations reproduce CDW phase transitions, the relation between the geometric metric and the actual charge-density wave should be established, e.g., by computing electronic structure or charge localization on representative snapshots, or by reframing the conclusions as referring to the lattice distortion only.","section":"Section II.C"},{"comment":"The hysteresis and multi-domain freezing are central conclusions, but the protocol uses 50 ps equilibration at each temperature and sequentially initializes each temperature from the previous final frame. No convergence tests with longer equilibration, independent initial conditions, or different supercell sizes are reported. The persistent multi-domain state at low temperature could result from slow kinetics in the 2028-atom cell rather than from a thermodynamically stable or long-lived physical state. The paper does acknowledge that the multi-domain structure persists 'at the cooling conditions considered here,' but this caveat does not fully address the concern because the claim is presented as reproducing experimental observations. Additional tests of equilibration time and initial-condition dependence would materially strengthen the hysteresis and domain-persistence claims.","section":"Section II.C, Fig. 4"}],"minor_comments":[{"comment":"The model name is spelled inconsistently ('UMA-s-1p1' vs. 'UMA s-1p1'). Please standardize.","section":"Global"},{"comment":"The abstract and Fig. 4 use 'pink line' and 'blue line' for heating and cooling; many readers may see grayscale printouts. Use distinct line styles or symbols as well.","section":"Abstract / Fig. 4"},{"comment":"The phrase 'part of Ta atoms' should be 'fraction of Ta atoms' or similar.","section":"Section III"},{"comment":"For TDEP phonons, the text states IFCs are extracted for the 'NM cell structure' from MD trajectories. At low temperatures the simulated system is in the CCDW state; the procedure for mapping those trajectories onto the NM cell should be described more explicitly to avoid ambiguity.","section":"Section II.D"},{"comment":"Reference [32] is cited as arXiv:2506.23971v2; please update to the published or final version if available. Reference [20] appears to be a journal article; please provide complete details.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely and potentially high-impact use of universal MLIPs for finite-temperature CDW simulations. The main concern is not circularity — the potential was pretrained and the MD outcomes are emergent — but validation coverage: the benchmark set does not include the intermediate and boundary configurations that carry the qualitative conclusions. I would be comfortable with acceptance after the authors either supply additional benchmark data for those configurations, or soften the claims to match what the current evidence supports. The geometric SoD metric vs. electronic CDW issue also needs to be addressed or explicitly delimited."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth your time. This is the first paper in its cited set to take a universal, zero-shot MLIP (UMA-s-1p1, no fine-tuning) and run 2028-atom MD through the 1T-TaS2 CDW sequence: SoD order melts near 290 K on heating, an NCCDW-like coexistence appears, the hexagonal phase is restored by ~500 K, cooling shows ~30 K hysteresis, and the layer freezes into persistent α/β chiral domains. That is a genuinely new demonstration — earlier MLIP-CDW studies on NbSe2 used system-specific training or nonthermal tuning. The paper also does a lot right: it benchmarks two uMLIPs against PBE displacement energies, uses independent metrics (Ta–Ta g(r) splitting, SoD fraction, TDEP soft-mode stabilization) that tell a consistent story, deposits analysis scripts and trajectory videos on Figshare, and states its limitations about quantitative Tc, cutoff radii, and missing SOC/magnetism.\n\nThe central weakness is the error budget. The benchmark reports MAE 0.048 eV/f.u. on random displacements from the two endpoint structures, while the NM–CCDW enthalpy difference is 0.027 eV/f.u. So the surrogate's stated error exceeds the energy scale that encodes relative phase stability. The MD results that matter — the NCCDW-like boundaries, the ~30 K hysteresis, the α/β domain walls that freeze in — come from configurations that are never benchmarked: partially formed SoD clusters, interdomain compressed bonds, transient ICDW-like motifs. That is a validation gap, not an internal inconsistency. The authors reproduce ΔH for the relaxed endpoints, which is encouraging, but two endpoints do not certify the energy surface along the transition paths. Second, the hysteresis and domain-multiplication narrative rests on single heating and cooling trajectories with no error bars; the SoD order parameter is a hand-thresholded geometric count, not an electronic order parameter. Third, the abstract's phrase 'reproduce the experimentally observed phase transition sequence' is stronger than the authors' own limitation statement warrants, especially since monolayer-specific experimental Tc values are not pinned down.\n\nNone of this kills the paper. The internal logic is coherent, the authors are honest, and the qualitative physics is plausible. But it is a proof-of-concept, not a definitive quantitative study.\n\nWho should read it: anyone working on MLIPs for correlated TMDCs, or on finite-T CDW simulations. It deserves serious peer review. The referee should push on the error-budget issue and ask for at least one repeated trajectory or an electronic-structure check of a few domain-wall snapshots. I'd send it to review with expectations of heavy revision.","headline":"First zero-shot universal MLIP to drive a finite-T CDW transition in 1T-TaS2; the physics story is coherent, but the surrogate's error budget leaves the domain/hysteresis claims on a validation gap.","tokens_in":12034,"tokens_out":3696,"would_cite":true,"duration_ms":31664,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A universal machine-learning interatomic potential, benchmarked only against density-functional displacement energies, reproduces the full heating–cooling sequence of charge-density-wave phases in monolayer 1T-TaS2 in 2028-atom molecular dy","keywords":["charge density waves","monolayer 1T-TaS2","universal machine learning interatomic potential","molecular dynamics","Star-of-David distortion","thermal hysteresis","temperature-dependent effective potential","CDW chirality domains"],"falsifier":"Compute UMA-s-1p1 energies for a set of domain-wall- and NCCDW-containing configurations drawn from the MD trajectories and compare against DFT energies for the same configurations; if the mean error in these unbenchmarked environments approaches or exceeds the 0.027 eV/f.u. NM-CCDW enthalpy difference, the simulated transition sequence, hysteresis width, and multi-domain freezing cannot be trusted as intrinsic lattice physics.","tokens_in":10885,"feed_emoji":"⚛️","tokens_out":4309,"duration_ms":35595,"temperature":0.7,"pith_summary":"The paper claims that a universal machine-learning interatomic potential, benchmarked only against DFT displacement energies, is accurate enough to drive 2028-atom molecular dynamics that reproduce the full heating sequence of charge-density-wave phases in monolayer 1T-TaS2—from the low-temperature Star-of-David commensurate phase through the nearly commensurate and incommensurate states to the high-temperature undistorted hexagonal metal. It further claims that cooling from high temperature yields a thermal hysteresis of roughly 30 K and freezes the system into a multi-domain state in which the two mirror-related CDW chiralities nucleate independently and persist to low temperature. Temperature-dependent effective potential phonons extracted from the trajectories show the CDW soft modes becoming dynamically stable as temperature rises, offering a vibrational signature of the transitions. A sympathetic reader would care because these transitions control transport switching in 1T-TaS2 devices, and the result suggests that zero-shot universal potentials can access finite-temperature phenomena previously out of reach for first-principles MD.","feed_headline":"Universal model reproduces TaS2 CDW phase sequence","feed_subtitle":"Zero-shot machine learning potential, benchmarked against DFT, runs 2,028-atom simulations capturing hysteresis and chiral domains.","key_machinery":"The central object is the Star-of-David (SoD) structural motif—13 Ta atoms contracting around a central Ta—used both as the physical fingerprint of CDW order and as a quantitative order parameter. The argument is carried by a geometric SoD-counting metric applied to time-averaged MD trajectories: a Ta atom is a SoD center if its six nearest neighbors all lie within 3.23 Å (temperature-adjusted), its second shell within 5.8 Å contains at least three Ta atoms, and at least three of the first-shell neighbours show the characteristic short-long bond pattern. This metric converts a 2028-atom trajectory into a number that tracks the fraction of Ta atoms in CDW clusters through the phase sequence.","core_discovery":"On the paper's own terms: a pretrained universal MLIP (UMA-s-1p1), selected by benchmarking against DFT displacement energies for the normal-metallic and commensurate CDW phases, is used in molecular dynamics on 2028-atom supercells of monolayer 1T-TaS2. Counting Ta atoms that belong to Star-of-David clusters yields a quantitative order parameter that drops sharply near 290 K on heating, passes through a nearly commensurate state of coexisting ordered and disordered regions, then an incommensurate-like state of transient clusters, and reaches zero in the hexagonal phase by about 500 K. Cooling reverses the transition about 30 K lower, and the low-temperature state is a multi-domain texture o","pith_inferences":["If the potential's accuracy generalizes to unbenchmarked domain-wall and mixed-state configurations, the same approach could be applied to other strongly coupled CDW materials (e.g., 1T-TaSe2, NbTe2) to map finite-temperature phase diagrams at near-first-principles cost.","The independence of α and β nucleation suggests a route to study chirality-controlled domains as bit-like objects; strain or defect pinning, which the paper lists as beyond scope, might be used to engineer domain sizes in simulations.","A direct test: run the same heating-cooling protocol with a second universal MLIP that passes the same flat-benchmark; if the hysteresis width or multi-domain persistence changes qualitatively, the current results may be potential-specific rather than material-intrinsic.","The SoD-counting metric could be applied to experimental diffuse scattering or STM topographies to infer domain fractions, connecting simulation order parameters to measurements."],"forward_implications":["A single universal potential can resolve the experimentally known CDW phase sequence in 1T-TaS2 at 2028-atom scale without material-specific training.","The computed ~290 K heating transition and ~30 K hysteresis match phenomenological observations, suggesting such simulations can locate approximate transition temperatures in CDW monolayers.","The persistence of α/β chiral domains on cooling provides a structural explanation for the suppressed long-range CCDW order reported in monolayer crystals.","TDEP phonons show that the high-temperature hexagonal phase is stabilized by anharmonic fluctuations, not by static harmonic stability.","The geometric SoD-counting metric can serve as a general order parameter for mixed CDW states in other TMDC monolayers."],"fun_headline_variants":["Universal MLIP captures TaS2 CDW phase sequence","Large-scale MD with universal potential reveals TaS2 CDW hysteresis","Zero-shot MLIP benchmarks and reproduces TaS2 CDW transitions","Machine-learned potential models TaS2 CDW phases and domains","Universal interatomic potential simulates TaS2 CDW order"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that UMA-s-1p1's energy surface is accurate in exactly the configurations that decide the results—intermediate NCCDW-like states, domain boundaries, and the nucleation events during heating and cooling—even though the published benchmark samples only small random displacements from the two end phases.","fun_headline_variants_meta":{"raw":{"variants":["Universal MLIP captures TaS2 CDW phase sequence","Large-scale MD with universal potential reveals TaS2 CDW hysteresis","Zero-shot MLIP benchmarks and reproduces TaS2 CDW transitions","Machine-learned potential models TaS2 CDW phases and domains","Universal interatomic potential simulates TaS2 CDW order"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000213,"raw_usage":{"total_tokens":1279,"prompt_tokens":785,"completion_tokens":494,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":415}},"tokens_in":529,"tokens_out":494,"duration_ms":4717,"temperature":1.0,"reasoning_tokens":415,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T05:08:27.021377+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute UMA-s-1p1 energies for a set of domain-wall- and NCCDW-containing configurations drawn from the MD trajectories and compare against DFT energies for the same configurations; if the mean error in these unbenchmarked environments approaches or exceeds the 0.027 eV/f.u. NM-CCDW enthalpy difference, the simulated transition sequence, hysteresis width, and multi-domain freezing cannot be trusted as intrinsic lattice physics.","supporting_citations":[],"review_version":1}