{"id":"561c370d-c28a-4ffe-a3ea-a5f5188b80ea","arxiv_id":"2411.14289","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Molecular dynamics with a machine-learned HSE06-trained potential predicts that BaZrS3 transforms from orthorhombic Pnma to tetragonal I4/mcm at 610 K and then to cubic Pm-3m at 880 K at zero pressure.","lead":"A computer simulation of BaZrS3, a lead-free solar and thermoelectric material, predicts that it switches between three crystal shapes as it is heated: at 610 K and again at 880 K. Knowing these shape changes matters because they could alter how the material performs in devices and how it forms during high-temperature synthesis.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Machine-learned potential resolution at the Pnma/I4/mcm crossing is unquantified and could shift the 610 K transition by hundreds of kelvin","rationale":"The paper is a solid computational study with a physically reasonable phase sequence; credit is due for the group-subgroup analysis, mode projections, thermodynamic integration, and public data and code. Those strengths do not, however, constrain the scale of free-energy difference that actually sets Tc. The 0 K gap (4.7 meV/atom) and latent heat (about 1 meV/atom) are both small enough that the model's 1.8 meV/atom training RMSE is a first-order concern, and the manuscript provides no convergence or uncertainty analysis for the TI crossing. The reader's weakest assumption already identified this NEP-resolution issue along with the Cmcm question; my primary attack is the same, so agreement is 'agree'. Cmcm is secondary but worth noting because the 'no other phases' claim depends on it. A DFT reweighting or delta-correction test would turn the qualitative concern into a quantitative bound. The original conditional verdict remains appropriate because the concern does not disprove the sequence; it argues for adding uncertainty quantification and a finite-temperature Cmcm check before the quantitative boundaries are taken as established.","tokens_in":24282,"tokens_out":6173,"duration_ms":61301,"concrete_test":"Run free-energy perturbation on top of the existing NEP TI simulation: draw roughly 100 statistically independent snapshots from each side at 580–640 K and 0 Pa, evaluate single-point HSE06 energies with the same settings as the training data, and estimate the DFT-level correction to ΔG_Pnma→I4/mcm at 610 K. If the corrected crossing temperature lies outside 610 ± 50 K, the reported transition temperature is not robust to the NEP approximation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"At 0 Pa, the central quantitative prediction is the 610 K Pnma-to-I4/mcm first-order transition. The load-bearing premise is that the NEP resolves the free-energy difference between these phases at 600–650 K. The 0 K Pnma–I4/mcm formation-energy gap is 4.7 meV/atom (Fig. 1c), only about 2.6 times the NEP energy RMSE of 1.8 meV/atom, and the latent heat at the transition is only about 1 meV/atom (Fig. 2d). The crossing is therefore set by a delicate enthalpy/entropy compensation: with ΔS ≈ 1/610 meV/K/atom, an uncorrected 1 meV/atom energy error shifts Tc by roughly 600 K. No uncertainty, training-set resampling, TI quadrature/finite-size check, or DFT single-point correction is reported for the 610 K (or 880 K) value. A related gap concerns Cmcm, observed as a minority phase at 570–770 K by Jaiswal et al.; the paper excludes it from 0 K dynamic instability alone, without computing its finite-temperature free energy. The qualitative Pnma→I4/mcm→Pm-3m sequence is plausible and independently supported by mode projections and XRD/Raman comparison, but the specific 610 K value—and the claim that no other phase intrudes—is not yet quantitatively secured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs a neuroevolution potential (NEP) trained on HSE06 hybrid DFT data for BaZrS3, then uses long NPT molecular dynamics simulations (40,960 atoms, up to 200 ns) and thermodynamic integration with an Einstein-crystal reference to map the finite-temperature phase behavior. At zero pressure the authors report a first-order orthorhombic Pnma-to-tetragonal I4/mcm transition at 610 K and a second-order I4/mcm-to-cubic Pm-3m transition at 880 K, build a pressure-temperature phase diagram from -4 to 10 GPa, and validate the model against published XRD and Raman data as well as a predicted temperature-dependent XRD pattern. The key claim is that this is the first report of these tilt-driven transitions in BaZrS3, with implications for high-temperature synthesis and thermoelectric applications.","tokens_in":24474,"tokens_out":4652,"duration_ms":40640,"significance":"If the quantitative predictions hold, the paper resolves conflicting experimental reports (Raman vs XRD) and provides the first clear sequence of tilt-driven phase transitions in BaZrS3, which is the most studied chalcogenide perovskite. The work is strengthened by the systematic enumeration of all Glazer tilt structures, the use of mode projections to identify phases in MD, the comparison with multiple experimental observables (XRD, Raman, lattice parameters, heat capacity), and the open availability of the NEP model, raw DFT data, and analysis code. The qualitative Pnma-to-I4/mcm-to-Pm-3m sequence is internally consistent with group-subgroup relations and is supported by the latent heat signature and mode amplitudes. However, the quantitative transition temperatures, especially the 610 K first-order transition, depend on the NEP resolving free-energy differences of order 1 meV/atom against a training RMSE of 1.8 meV/atom, and no uncertainty or convergence analysis is provided; the treatment of the experimentally suggested Cmcm phase is also incomplete.","major_comments":[{"comment":"The 610 K Pnma-to-I4/mcm transition temperature is not accompanied by any uncertainty or convergence analysis, which is load-bearing because the free-energy difference being resolved is very small. The 0 K Pnma-I4/mcm formation-energy difference is 4.7 meV/atom (Fig. 1c) and the latent heat is about 1 meV/atom (Fig. 2d), while the NEP energy RMSE is 1.8 meV/atom; with ΔS ≈ 0.0016 meV/K/atom, a 1 meV/atom error shifts Tc by roughly 600 K. The manuscript should report TI convergence with respect to the λ quadrature, the Einstein-crystal spring constant, and system size, and should provide an estimate of NEP error propagation (for example, via training-set resampling or DFT single-point energies on MD snapshots near the crossing).","section":"Methods (Thermodynamic integration) and Fig. 2d"},{"comment":"The Cmcm phase, which Jaiswal et al. observed as a minority phase at 570-770 K, is excluded from the phase diagram based only on its 0 K dynamic instability and relaxation to I4/mcm. Because the energy difference between the symmetry-constrained Cmcm and the relaxed I4/mcm is as small as 0.2 meV (per the SI), entropic stabilization at finite temperature cannot be ruled out without computing the Cmcm free energy at the relevant temperatures. The conclusion that no other phase intrudes before melting requires such a calculation or, at minimum, a free-energy comparison of Cmcm against I4/mcm and Pnma in the 500-900 K range.","section":"Discussion of Cmcm (after Fig. 3) and Table S1"},{"comment":"The second-order I4/mcm-to-Pm-3m transition temperature of 880 K is taken from heating simulations only, with no free-energy calculation, finite-size analysis, or error estimate. Since the heat-capacity peak is broad (Fig. 2e), the extracted transition temperature carries an unspecified systematic uncertainty that should be quantified by, for example, comparing different heating rates, cell sizes, or by locating the free-energy crossing.","section":"Fig. 3 caption and Fig. 2"}],"minor_comments":[{"comment":"The abstract quotes 610 K for the first transition while the heating simulation described in the text and Fig. 2 gives 650 K; the manuscript should explicitly state that 610 K is the thermodynamic-integration result and explain the difference with respect to the raw MD transition temperature.","section":"Abstract vs. Fig. 2 and text"},{"comment":"The sentence 'In contrast, P 4/mbm is a subgroup of the Pm 3m phase, so can be accessed through a second-order transition' appears to reference the wrong space group; the relevant subgroup for the I4/mcm-to-Pm-3m transition is I4/mcm itself, and the sentence should be corrected.","section":"Paragraph after Fig. 2"},{"comment":"Several references in the SI Methods are unresolved placeholders ('Ref. ?', '?', 'citeglazer1972classification'), and the SI should be completed with proper citations and bibliographic details before publication.","section":"Supplementary Information Methods"},{"comment":"The name 'Jaiswel' appears once in the main text; this is a typo for 'Jaiswal', matching the cited reference.","section":"Main text, Raman discussion"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a good fit for the journal and the qualitative phase-transition sequence is likely correct, but the quantitative claims are not yet secured because the NEP's ability to resolve free-energy differences at the first-order transition is unquantified and the Cmcm issue is not addressed at finite temperature. The requested additions (TI convergence, finite-size checks, uncertainty estimates, Cmcm free energy) are feasible within the manuscript's scope. I also note the SI is incomplete in its current form and should be fixed during revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a read: this is the first complete Pnma-to-I4/mcm-to-Pm-3m sequence for BaZrS3, and I think the qualitative picture is right. The specific 610 K number, though, is less secure than the abstract suggests.\n\nThe paper's real contribution is assembling the full Glazer-tilt map, showing Pnma as the ground state and I4/mcm only 4.7 meV/atom above it, then demonstrating with 40,960-atom MD that heating goes Pnma to I4/mcm (first order) and I4/mcm to Pm-3m (second order), with thermodynamic integration giving the free-energy crossing at 610 K. The second-order transition at 880 K is consistent with group-subgroup analysis. They openly release the NEP model, DFT data, and code, and they validate against experimental XRD and Raman from Bystricky and Jaiswal. The predicted temperature-dependent XRD pattern is a useful falsifiable output. Those are real strengths.\n\nThe weak point is quantitative. The 0 K Pnma-I4/mcm enthalpy difference is 4.7 meV/atom against a training RMSE of 1.8 meV/atom, and the latent heat is roughly 1 meV/atom. So the crossing at 610 K rests on a delicate enthalpy/entropy compensation, and no uncertainty estimate, resampling, or DFT single-point check is reported. The harmonic approximation gives 243 K; the anharmonic TI gives 610 K, so model dependence is large. Also, Cmcm is excluded from 0 K dynamic instability alone; the experimental minority phase at 570-770 K deserves a finite-temperature free-energy comparison. The XRD validation against I41/acd rather than I4/mcm is acknowledged but could be tighter. These are not fatal—mode projections, latent heat, and symmetry arguments all point the same way—but the authors should be asked to bound the error bars on Tc.\n\nThis deserves peer review. It is a well-executed computational study with open data and a new phase diagram for an important material. The revisions should focus on adding convergence/uncertainty estimates and a finite-T treatment of Cmcm. I would send it to a good condensed-matter/materials journal with a referee who knows ML potentials, and push for the UQ before publication.","headline":"First complete tilt sequence for BaZrS3, with a credible phase diagram; treat the 610 K first-order boundary as qualitatively right but numerically unquantified.","tokens_in":25113,"tokens_out":3765,"would_cite":true,"duration_ms":32898,"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":"BaZrS3, the most-studied lead-free chalcogenide perovskite, is predicted to transform from orthorhombic to tetragonal at 610 K and to cubic at 880 K.","keywords":["BaZrS3","chalcogenide perovskite","octahedral tilting","phase transition","machine-learned interatomic potential","molecular dynamics","thermodynamic integration","X-ray diffraction"],"falsifier":"Heat BaZrS3 in an inert atmosphere and collect synchrotron X-ray diffraction every few kelvin from 500 to 900 K: if no first-order discontinuity appears near 610 K, or if Cmcm reflections appear as a stable phase between 570 and 770 K, the predicted sequence is wrong. A purely computational falsifier is to recompute the Pnma and I4/mcm Gibbs free energies at 610 K with a potential whose training error is well below 1 meV/atom and see whether the crossing survives.","tokens_in":24011,"feed_emoji":"⚛️","tokens_out":9017,"duration_ms":77526,"temperature":0.7,"pith_summary":"BaZrS3, the most-studied lead-free chalcogenide perovskite for photovoltaics and thermoelectrics, is normally characterized at room temperature, where it is orthorhombic. This paper predicts what happens when it gets hot: at zero pressure it switches to a tetragonal I4/mcm structure at 610 K in a first-order transition, and then to the cubic Pm-3m structure at 880 K in a second-order transition. The predictions come from a machine-learned interatomic potential trained on hybrid density-functional calculations, used in molecular-dynamics simulations and thermodynamic-integration free energies. Because synthesis and thermoelectric operation both reach above 610 K, the paper implies that high-temperature samples and devices involve tetragonal or cubic BaZrS3, not the room-temperature phase, and that pressure widens the tetragonal window.","feed_headline":"BaZrS3 leaves its room-temperature structure at 610 K","feed_subtitle":"Machine-learned atomistic simulations map BaZrS3's tilt-driven phases; the tetragonal phase widens under pressure.","key_machinery":"The argument runs on four connected tools. A neuroevolution potential (a machine-learned interatomic potential fit to hybrid DFT data) supplies energies, forces, and stresses for molecular dynamics; projections of atomic displacements onto the M- and R-point tilt eigenvectors of the cubic cell classify each simulated snapshot as Pnma, I4/mcm, or Pm-3m; thermodynamic integration with an Einstein-crystal reference gives the free energies that place the first-order transition at 610 K; and the group–subgroup graph of the 15 Glazer tilt patterns says which transitions can be second order and which must be first order.","core_discovery":"The central claim is that BaZrS3 follows the standard perovskite tilt sequence with heating: orthorhombic Pnma (a+b−b− tilts) converts to tetragonal I4/mcm (a0a0c−) at 610 K, and I4/mcm converts to cubic Pm-3m (a0a0a0) at 880 K. The first transition is discontinuous, with a latent heat of about 1 meV per atom, and the second is continuous; both characters match what the group–subgroup relations allow. The paper further claims that the tetragonal phase is stable over a wider temperature range at higher pressure, with the first-order transition temperature saturating near 690 K above 4 GPa, and it supplies a temperature-dependent X-ray diffraction pattern that experimentalists can use to identify the phases. No other tilt phases appear between the Pnma ground state and the cubic phase.","pith_inferences":["Because the 0 K energy gap between Pnma and I4/mcm (4.7 meV/atom) is only about 2.5 times the potential's training error (1.8 meV/atom), a reasonable bound on the predicted 610 K crossing is several tens of kelvin; an ensemble of potentials or an explicit uncertainty estimate would test that.","The most experimentally contested point is whether Cmcm is a genuine intermediate phase: the paper excludes it from 0 K stability, but an explicit finite-temperature free-energy calculation for Cmcm would be a sharper test than the current stability argument.","The same workflow can be transferred directly to BaHfS3 and other chalcogenide perovskites, whose high-temperature phase sequences are not yet mapped.","A-site off-centering visible in the X-point diffraction peaks suggests local polar distortions may coexist with tilts below 610 K, which could influence carrier mobility or recombination even in the 'nonpolar' Pnma phase."],"forward_implications":["Samples grown above about 850 K, the typical synthesis temperature, will pass through the 610 K transition on cooling and may contain mixtures of tetragonal and orthorhombic polymorphs.","Thermoelectric devices operating in the 400–1100 K range will cross both transitions, so calculations of transport, thermal conductivity, and band gaps should use the I4/mcm or cubic structures above 610 K.","Applying pressure or compressive substrate strain stabilizes the tetragonal phase over a wider temperature interval, while tensile strain (negative pressure) favors the higher-symmetry phases, offering a handle for interface engineering.","The predicted XRD pattern, including the R-mode superlattice peak near 29° in tetragonal BaZrS3 and M-mode peaks near 27° and 33° in Pnma, provides a direct fingerprint for identifying phases in high-temperature experiments."],"supporting_citations":[{"why":"Experimental XRD patterns at 303 K and 923 K used to validate the predicted Pnma and I4/mcm diffraction signatures.","marker":"31"},{"why":"Multimodal experimental study that found Pnma below 570 K, I4/mcm above 770 K, and a Cmcm minority phase; provides the key experimental benchmark.","marker":"32"},{"why":"Machine-learned potential molecular-dynamics study of halide-perovskite phase transitions; supplies the simulation methodology and latent-heat comparison.","marker":"35"},{"why":"Thermodynamic-integration study of transition characters in halide perovskites; supports the interpretation of hysteresis and first-order barriers.","marker":"36"},{"why":"GPUMD package implementing the NEP model used for all molecular dynamics and free-energy calculations.","marker":"41"},{"why":"Glazer classification enumerating the 15 octahedral-tilt patterns that define the candidate phase space.","marker":"44"},{"why":"Howard–Stokes group-subgroup analysis and Landau rules used to classify allowed first- and second-order transitions.","marker":"45"},{"why":"HSE06 hybrid-functional DFT reference data for the 1187 training structures that define the accuracy of the machine-learned potential.","marker":"47"},{"why":"Phonon mode-projection method used to assign MD snapshots to Pnma, I4/mcm, or Pm-3m by M- and R-mode amplitudes.","marker":"49"},{"why":"Frenkel–Ladd thermodynamic integration with an Einstein reference crystal used to compute free energies and locate the 610 K transition.","marker":"50"}],"fun_headline_variants":["BaZrS3 undergoes two tilt-driven phase transitions on heating","BaZrS3 jumps to tetragonal at 610 K, then to cubic at 880 K","Machine-learned potential maps BaZrS3 phase boundaries and pressure effect","Pressure widens BaZrS3's tetragonal phase, simulations show","BaZrS3's tilts: discontinuous jump at 610 K, continuous at 880 K"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the trained potential resolves the few-meV-per-atom free-energy differences between competing tilt arrangements at 600–900 K, and that the Cmcm arrangement, a minority phase in experiments, has no finite-temperature stability even though only its 0 K instability was checked.","fun_headline_variants_meta":{"raw":{"variants":["BaZrS3 undergoes two tilt-driven phase transitions on heating","BaZrS3 jumps to tetragonal at 610 K, then to cubic at 880 K","Machine-learned potential maps BaZrS3 phase boundaries and pressure effect","Pressure widens BaZrS3's tetragonal phase, simulations show","BaZrS3's tilts: discontinuous jump at 610 K, continuous at 880 K"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000901,"raw_usage":{"total_tokens":3886,"prompt_tokens":963,"completion_tokens":2923,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":579,"completion_tokens_details":{"reasoning_tokens":2829}},"tokens_in":579,"tokens_out":2923,"duration_ms":18162,"temperature":1.0,"reasoning_tokens":2829,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:20:49.217721+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Heat BaZrS3 in an inert atmosphere and collect synchrotron X-ray diffraction every few kelvin from 500 to 900 K: if no first-order discontinuity appears near 610 K, or if Cmcm reflections appear as a stable phase between 570 and 770 K, the predicted sequence is wrong. A purely computational falsifier is to recompute the Pnma and I4/mcm Gibbs free energies at 610 K with a potential whose training error is well below 1 meV/atom and see whether the crossing survives.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Experimental XRD patterns at 303 K and 923 K used to validate the predicted Pnma and I4/mcm diffraction signatures."},{"cited_title":"A.; Lekina, Y.; Kamonsuangkasem, K.; Tomm, Y.; Wei, F.; White, T","cited_arxiv_id":null,"evidence_quote":"Multimodal experimental study that found Pnma below 570 K, I4/mcm above 770 K, and a Cmcm minority phase; provides the key experimental benchmark."},{"cited_title":"Differing vibrational properties of halide and chalcogenide perovskite semiconductors and impact on optoelectronic performance","cited_arxiv_id":null,"evidence_quote":"Machine-learned potential molecular-dynamics study of halide-perovskite phase transitions; supplies the simulation methodology and latent-heat comparison."},{"cited_title":"M.; Wiktor, J.; Erhart, P","cited_arxiv_id":null,"evidence_quote":"Thermodynamic-integration study of transition characters in halide perovskites; supports the interpretation of hysteresis and first-order barriers."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"GPUMD package implementing the NEP model used for all molecular dynamics and free-energy calculations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Glazer classification enumerating the 15 octahedral-tilt patterns that define the candidate phase space."},{"cited_title":"calorine: A Python package for constructing and sampling neuroevolution potential models","cited_arxiv_id":null,"evidence_quote":"Howard–Stokes group-subgroup analysis and Landau rules used to classify allowed first- and second-order transitions."},{"cited_title":"J.; Stokes, H","cited_arxiv_id":null,"evidence_quote":"HSE06 hybrid-functional DFT reference data for the 1187 training structures that define the accuracy of the machine-learned potential."},{"cited_title":"V.; Vydrov, O","cited_arxiv_id":null,"evidence_quote":"Phonon mode-projection method used to assign MD snapshots to Pnma, I4/mcm, or Pm-3m by M- and R-mode amplitudes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Frenkel–Ladd thermodynamic integration with an Einstein reference crystal used to compute free energies and locate the 610 K transition."}],"review_version":1}