{"id":"a666faac-3ad8-47c7-a4db-0c6f246df194","arxiv_id":"2507.07926","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Machine-learned force fields predict that methylammonium suppresses the beta-to-gamma transition in mixed lead-halide perovskites unless the organic cations rearrange, while formamidinium stabilizes a cubic low-temperature phase with a+a+a+ tilts.","lead":"This paper simulates how lead-halide perovskite crystals, a leading solar-cell material, change their internal tilting patterns with temperature and with mixing iodine and bromine. It finds that the organic molecule methylammonium blocks a particular low-temperature structural change unless the molecules rotate together, and that separated iodine-rich and bromine-rich regions create conflicting tilts that resist uniform transformation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The beta-to-gamma 'forbidden' claim rests on random-init, 0.2 K/ps cooling runs; the authors' own gamma-aligned runs show strong initial-condition dependence, so the unqualified abstract claim overstates a finite-time kinetic result.","rationale":"The reader's weakest assumption was transferability of the MACE potentials from small training cells to large production supercells, including the acknowledged systematic softening. That is a valid concern, but the single most load-bearing issue for the headline claim is different: the beta-to-gamma 'forbidden' conclusion is a kinetic, finite-time, initial-condition-dependent statement, and the paper's own evidence shows that gamma can be obtained when MA molecules are initialized in gamma-compatible orientations. This makes the unqualified abstract wording an overclaim, and it means the central assertion rests on absence of nucleation in short MD runs rather than on a measured barrier. The proposed test would directly quantify the free-energy barrier and thereby distinguish a true kinetic suppression from a simulation-rate artifact. The miscibility-gap polynomial fit and missing artifact release are secondary; they do not affect the core beta-to-gamma claim as directly. On balance, the work remains valuable and internally largely consistent, but the conditionality already assigned by the reader is appropriate, and no verdict change is needed.","tokens_in":19634,"tokens_out":6495,"duration_ms":70786,"concrete_test":"Using the released MACE potential, compute the free-energy profile (umbrella sampling or metadynamics) along a collective variable that measures MA molecular orientation relative to the beta-phase orientation at 150 K in a 4x4x4 MAPbI3 cell. If the barrier separating beta and gamma orientation basins is below about 0.2 eV, the transition should occur on accessible timescales and the 'forbidden' claim is a rate artifact; if the barrier exceeds about 0.5 eV and is consistent with r2SCAN DFT, the claim is supported. Complement this with 0.02 K/ps cooling runs in a 10x10x10 cell to check whether any run nucleates gamma from a random beta-phase starting point.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that MA+ 'forbids' the beta-to-gamma transition in MAPbX3 is supported only by constant-rate cooling at 0.2 K/ps starting from beta-phase structures with randomly rotated MA molecules (Section III.B-III.C). The authors state that 'when the molecules in the initial structures are aligned as in the pattern found in equilibrated gamma phase, the gamma phase can be found in a wider window of temperatures and halide compositions' (Section III.C, Fig. 4). Thus the transition is not intrinsically forbidden; it is suppressed for one initial-condition protocol on a nanosecond timescale. The abstract drops the 'finite cell and time' qualifier that the conclusion uses, turning a kinetic finite-time observation into a categorical statement. No free-energy barrier or nucleation rate is computed, so the inference that nucleation fails because gamma-orientations are not a subset of beta-orientations is an absence-of-events argument from eight runs, not a demonstrated barrier. Real polycrystalline samples contain surfaces, grain boundaries, and defects that can provide heterogeneous nucleation sites absent from the pristine 10x10x10 supercells; experimental MAPbI3 does undergo the tetragonal-to-orthorhombic transition near 160 K. The authors' own admission that the phase diagram is incomplete because of randomly rotated initial MA reinforces the concern, and the acknowledged systematic softening of the MACE potential could mis-set the rotational barriers that this conclusion depends on.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript trains three MACE machine-learning force fields on on-the-fly r2SCAN DFT data for CsPbX3, MAPbX3, and FAPbX3 and uses them in large-scale NpT molecular dynamics (up to ~13,700 atoms) with the PDynA toolkit to construct phase diagrams, extract octahedral tilt correlations, and analyze phase transformation pathways for pure and mixed-halide systems. The principal claims are that MA+ effectively forbids the beta-to-gamma transition in MAPbX3 because it requires extensive molecular reorientation and crystal rotation; that low-temperature FAPbX3 is best described as an Im-3 cubic phase with a+a+a+ tilts; and that halide segregation creates anomalous tilt modes that impede uniform transformations.","tokens_in":19857,"tokens_out":6390,"duration_ms":70687,"significance":"The paper is potentially significant for the halide-perovskite community: it demonstrates a practical workflow combining on-the-fly active learning, MACE potentials, and PDynA descriptors to reach length and time scales far beyond direct DFT, and it makes concrete, testable structural predictions, notably the Im-3 a+a+a+ phase for FAPbI3, which is compared with independent diffuse-scattering and neutron work. The local octahedral classification by halide configuration (Fig. 2) and the segregation-induced tilt analysis (Fig. 9) are valuable and go beyond bulk-averaged descriptors. However, the headline 'forbidden' transition is a finite-time kinetic observation, and the thermodynamic miscibility analysis is based on an overparameterized fit; these issues currently limit the strength of the conclusions that can be drawn.","major_comments":[{"comment":"The abstract states that MA+ effectively 'forbids' the beta-to-gamma transition in MAPbX3, but the body supports only a substantially weaker statement. Section III.C reports that when MA molecules are initially aligned as in the equilibrated gamma pattern, 'the gamma phase can be found in a wider window of temperatures and halide compositions' (Fig. 4, dashed line), and the conclusion itself adds the qualifier 'in a finite cell and time.' No free-energy barrier or nucleation rate is computed; the conclusion is an absence-of-events observation from eight constant-rate (0.2 K/ps) cooling runs on pristine supercells. The abstract should be qualified to state that beta-to-gamma is strongly suppressed under the simulated finite-cell, finite-time protocols, rather than categorically forbidden.","section":"Abstract and Section III.C"},{"comment":"All central results depend on three bespoke MACE potentials, yet Section V only states that a repository with full training data and force-field parameters 'will also be provided,' without a current link or DOI. Without these data and parameters, the RMSEs in Table S1 cannot be checked and the transferability from 2x2x2 training cells (40-96 atoms) to production cells of 10x10x10-14x14x14 (12,000-13,720 atoms) cannot be assessed. This is especially important because the authors acknowledge a systematic softening of universal ML potentials that already shifts transition temperatures below experiment (Section III.A); the beta-to-gamma barrier inference inherits this bias. The potentials and training data should be released and, ideally, additional validation of large-scale tilt behavior (e.g., against DFT or experimental transition temperatures for several compositions) should be provided.","section":"Section V and II.B"},{"comment":"The thermodynamic-stability analysis fits a fifth-order polynomial to seven enthalpy-of-mixing data points and then uses its second derivative to construct miscibility-gap and spinodal curves. A fifth-order polynomial has six coefficients, so the fit has essentially one degree of freedom; the resulting common-tangent construction is extremely sensitive to the polynomial form and no uncertainty or cross-validation is reported. The statement that 'the FA system possesses the highest stability among the three solid solutions' (Section III.C) therefore rests on a fragile numerical procedure. Either a lower-order fit with error estimates, a different thermodynamic model, or explicit validation against the cited DFT predictions is needed to support the quantitative stability ranking.","section":"Section III.C, Eq. 7 and Fig. 8"},{"comment":"The phase classification underlying Fig. 4 uses criteria of 'insignificant tilt angle or a TCP value near zero' without reporting the numerical thresholds. Because every phase assignment and transition temperature in the equilibrium phase diagrams is derived from this thresholding, the maps could change substantially under reasonable alternative cutoffs. The authors should state the thresholds and demonstrate that the reported phase boundaries are robust to them.","section":"Section III.C (phase identification)"}],"minor_comments":[{"comment":"The text contains a typo: 'structural descritor' should be 'structural descriptor.'","section":"Section II.B"},{"comment":"The caption contains a typo: 'crystal srtucture' should be 'crystal structure.'","section":"Fig. 6 caption"},{"comment":"The text 'force field paramaters' contains a typo; it should read 'force-field parameters.'","section":"Section V"},{"comment":"The reference tilt angles \\hat{\\theta}_i^j are said to be obtained from equilibration calculations, but the exact procedure (which time windows, which compositions, and how averages are taken) is not specified; please clarify.","section":"Equation (6)"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nRead this one if you work on halide perovskite phase behavior. The genuinely new content is the A-site cation mechanism: the paper argues with large-scale MD and PDynA tilt analysis that the MAPbX3 beta-to-gamma transition is suppressed because gamma-phase MA orientations are not reachable from beta-phase distributions in accessible cooling runs, whereas FA retains nearly the same orientation across phases and transforms more easily. That mechanistic contrast is well supported by orientation-distribution statistics, hydrogen–halide pdfs, and external kinetics and neutron references. The additional results—Im-3 a+a+a+ low-temperature FAPbI3, segregation-induced b- tilts and directional crystallization, and reduced tilt correlation lengths under halide mixing—are useful and mostly measured rather than asserted.\n\nThe workflow is coherent: on-the-fly r2SCAN data, MACE training with reported RMSEs, LAMMPS production runs, and PDynA-based phase classification. Credit also for the authors stating plainly that the predicted transitions are softened relative to experiment, that relative stabilities matter more than absolute temperatures, and that the MAPbX3 phase diagram is incomplete because initial MA orientations were random. The body text repeatedly qualifies the beta-to-gamma claim with finite cell and time.\n\nSoft spots, in order. First, the abstract says MA+ effectively forbids the beta-to-gamma transition, without the finite-cell, finite-time qualifiers. That is an overstatement: the evidence is absence of events in a handful of constant-rate cooling runs under one initial-condition protocol. The authors themselves show that gamma-aligned initial molecules produce gamma phase over a wider window, so the result is kinetic suppression under a specific protocol, not a demonstrated thermodynamic barrier. No free-energy barrier or nucleation rate is computed. Still, the mechanism is plausible and consistent with experiment, so this is an overreach in wording rather than a broken core. Second, training data and MACE parameters are promised but not yet released, so reproducibility currently rests on PDynA plus the text. Third, the miscibility-gap and spinodal curves come from an unquantified fifth-order polynomial fit to seven enthalpy points; that should be labeled a rough estimate. Fourth, transferability from small training cells to 12,000+ atom supercells is a real risk, though the acknowledged systematic softening makes the relative phase ordering the sensible reading.\n\nBottom line: this is a solid computational study with one headline overclaim and two fixable reproducibility gaps. It deserves a serious referee; I would send it out rather than desk-reject.","headline":"A solid perovskite MLFF study with a credible A-site mechanism; the headline 'forbids' overreaches beyond finite-time kinetics, but the paper deserves serious peer review.","tokens_in":20470,"tokens_out":2019,"would_cite":true,"duration_ms":24874,"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":"Methylammonium effectively forbids the low-temperature phase transition in MAPbX3.","keywords":["lead halide perovskites","phase transitions","machine learning force fields","octahedral tilting","methylammonium","halide segregation","molecular dynamics","phase diagrams"],"falsifier":"A direct test would be a free-energy calculation, not just a cooling trajectory: use umbrella sampling or metadynamics on the same potentials to compute the $\\beta\\to\\gamma$ free-energy barrier in MAPbX$_3$; if the barrier is only a few $k_{\\mathrm{B}}T$ at 150 K, then the arrest is a finite-time artifact. On the experimental side, anneal a MAPbI$_3$ single crystal just below the $\\beta$–$\\gamma$ boundary for hours while monitoring neutron or X-ray diffuse scattering and molecular orientation; appearance of the $\\gamma$-phase orientation pattern would refute the claim that the transition is effectively forbidden in accessible conditions.","tokens_in":19288,"feed_emoji":"🧊","tokens_out":10432,"duration_ms":105590,"temperature":0.7,"pith_summary":"Using machine-learned force fields trained on density-functional-theory data, the paper simulates cesium, methylammonium, and formamidinium lead halide perovskites with mixed iodine/bromine in supercells of 12,000 to 13,720 atoms and builds equilibrium phase diagrams plus 0.2 K/ps heating and cooling runs. The central claim is that methylammonium ($\\mathrm{MA}^+$) effectively forbids the $\\beta$-to-$\\gamma$ transition in MAPbX$_3$: the $\\gamma$-phase needs a molecular-orientation pattern and crystal rotation that the $\\beta$-phase cannot provide, so nucleation fails in a finite cell and time. A second claim assigns the debated low-temperature phase of FAPbX$_3$ to a cubic Im$\\bar{3}$ structure with $a^+a^+a^+$ octahedral tilts. The paper also shows that halide mixing shortens tilt correlation lengths and that segregated iodine-rich domains develop an anomalous $b^-$ tilt that lowers and directionally biases phase transitions. This matters because the same structural physics governs the thermal and photo-stability of perovskite solar cells.","feed_headline":"Methylammonium blocks a key phase change in lead halide perovskites","feed_subtitle":"Large-scale simulations show the organic cation's rotation prevents the beta-to-gamma transition.","key_machinery":"The argument runs on three coupled descriptors extracted from large-scale molecular dynamics. The first is the octahedral tilt field: tilting correlation polarity (TCP) along each principal axis, with values $-1$, $0$, $+1$ playing the role of Glazer superscripts, and a correlation-length tensor obtained by fitting the decay of tilt-tilt correlations to an exponential. The second is the molecular-orientation (MO) distribution of the organic cations, projected in a full octahedral-symmetry frame and in a no-symmetry frame, with a distribution variance that measures how concentrated the orientations are; the subset relations among $\\alpha$, $\\beta$, and $\\gamma$ orientation sets are what make a transition free, guided, or limited. The third is the segregation parameter $k_{\\mathrm{seg}}$, which counts how many bromine atoms surround each octahedron and quantifies the Monte Carlo-driven I/Br clustering. Together, these descriptors let the authors classify phases, locate transition temperatures from the crossing of angle-deviation curves, and attribute the MAPbX$_3$ arrest to molecular-orientation mismatch rather than to a bare static energy barrier.","core_discovery":"On the paper's own terms, the finding is that the A-site cation controls which symmetry-lowering pathways are kinetically accessible. For MAPbX$_3$, the $\\gamma$-phase orientation distribution of the $\\mathrm{MA}^+$ molecules is not a subset of the $\\alpha$- or $\\beta$-phase distributions, so the $\\beta$-to-$\\gamma$ transition demands extensive molecular rearrangement and crystal rotation; in the finite 12,000-atom cells and 0.2 K/ps cooling windows used here, it is effectively forbidden. The $\\alpha$-to-$\\beta$ transition remains easy because the $\\beta$ orientation set is a subset of the $\\alpha$ set. For FAPbX$_3$, the $\\mathrm{FA}^+$ molecules keep the same $\\langle100\\rangle$ preference across all phases, so phase changes are free and the low-temperature phase is best represented as Im$\\bar{3}$ with $a^+a^+a^+$ tilts, consistent with a recent diffuse-scattering experiment. Halide composition enters through tilt correlations: random I/Br mixing reduces the tilt correlation length below the linear interpolation of the pure endpoints, while partial segregation creates I-rich domains with unusually strong, anisotropic tilting and an anomalous $b^-$ mode that impedes coherent transformation. The paper further reports hysteresis between heating and cooling transitions, larger in mixed compositions, and $\\Delta G_{\\mathrm{mix}}$ estimates indicating FAPbX$_3$ is the most stable solid solution.","pith_inferences":["The orientation-subset criterion is a general screening rule: for any hybrid perovskite, compare the molecular orientation distributions of adjacent phases; when the lower-phase distribution is not a subset of the higher-phase one, expect kinetic arrest, regardless of the cation's identity. This could be tested cheaply on other organic cations such as aziridinium or ethylammonium.","Because the paper's own softening caveat shifts absolute transition temperatures, the 'forbidden' claim is best read as a statement about kinetics in finite systems, not thermodynamics; enhanced-sampling simulations or microsecond runs on the same potentials might still nucleate $\\gamma$ and would delimit how strong the barrier really is.","The finding that segregation predetermines crystallization orientation suggests a feedback loop in operating devices: photoinduced halide demixing not only creates trap states but also selects the crystal orientation of the low-temperature phase, which may in turn influence ion migration and further segregation. This is a testable hypothesis with in-situ X-ray diffraction during illumination.","The correlation-length metric could serve as a design proxy: compositions whose tilt correlation lengths are long and isotropic should transform more coherently, so screening candidates by tilt-correlation tensors may be a faster route to phase-stable alloys than enumerating full phase diagrams."],"forward_implications":["If $\\mathrm{MA}^+$ really forbids the $\\beta$-to-$\\gamma$ transition in finite crystals, then ordinary thermal cycling of MAPbI$_3$ leaves the material kinetically trapped in the tetragonal/cubic manifold, and the orthorhombic phase appears only with seeding, strain, or much slower cooling.","If FAPbX$_3$'s low-temperature phase is Im$\\bar{3}$ with $a^+a^+a^+$ tilts, then the FA system's transitions involve no molecular reorientation barrier, so its phase stability is governed almost entirely by octahedral tilting and is therefore most sensitive to force-field softening.","If random halide mixing shortens tilt correlation lengths below the endpoint interpolation, then mixed I/Br compositions intrinsically suppress the $\\alpha$-to-$\\beta$ transition temperature relative to the pure halides, which the free-energy of mixing data corroborate.","If segregated domains foster an anomalous $b^-$ tilt, then light- or temperature-induced halide segregation can change which crystallographic variant forms and even predetermine its orientation, coupling degradation directly to structural phase behavior."],"supporting_citations":[{"why":"Supplies the equivariant message-passing neural-network architecture used for the production-scale molecular dynamics.","marker":"[66]"},{"why":"Supplies the tilt-correlation polarity, correlation-length, and molecular-orientation descriptors used for phase classification.","marker":"[74]"},{"why":"Provides the experimental diffuse-scattering and neutron data identifying an Im-3 a+a+a+ phase that the low-temperature FAPbX3 assignment is checked against.","marker":"[35]"},{"why":"Describes the on-the-fly Bayesian-error data collection strategy used to build the training sets.","marker":"[63]"},{"why":"Gives DFT kinetics for the tetragonal-to-orthorhombic transformation of MAPbI3 that support the MA-rearrangement barrier interpretation.","marker":"[90]"},{"why":"Provides neutron-scattering measurements of methylammonium reorientation dynamics consistent with the required molecular rearrangement.","marker":"[91]"},{"why":"Documents the universal molecular reorientation behavior in hybrid lead iodide perovskites that underlies the FA orientation continuity.","marker":"[93]"},{"why":"Documents the systematic softening of universal machine-learning potentials that the authors invoke to explain their lowered transition temperatures.","marker":"[83]"},{"why":"Provides the group-theoretical analysis of octahedral tilt transitions that underpins the free/guided/limited pathway classification.","marker":"[87]"}],"fun_headline_variants":["Methylammonium forbids beta-to-gamma perovskite shift","Halide mixing alters tilt order in lead perovskites","Neural-network MD maps perovskite phase anomalies","Cation rotation controls perovskite phase transitions","FA+ allows phase changes; MA+ blocks them"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the three trained force fields remain accurate when taken from small 2×2×2 training cells and short on-the-fly trajectories to production supercells of 12,000–13,720 atoms run at 0.2 K/ps heating and cooling; the authors openly note a systematic softening that already shifts transition temperatures below experiment, so if the potentials invent or suppress tilt instabilities at scale, the phase diagrams and the 'forbidden' transition claim shift with them.","fun_headline_variants_meta":{"raw":{"variants":["Methylammonium forbids beta-to-gamma perovskite shift","Halide mixing alters tilt order in lead perovskites","Neural-network MD maps perovskite phase anomalies","Cation rotation controls perovskite phase transitions","FA+ allows phase changes; MA+ blocks them"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000779,"raw_usage":{"total_tokens":3525,"prompt_tokens":1110,"completion_tokens":2415,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":726,"completion_tokens_details":{"reasoning_tokens":2352}},"tokens_in":726,"tokens_out":2415,"duration_ms":21202,"temperature":1.0,"reasoning_tokens":2352,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:30:07.073119+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test would be a free-energy calculation, not just a cooling trajectory: use umbrella sampling or metadynamics on the same potentials to compute the $\\beta\\to\\gamma$ free-energy barrier in MAPbX$_3$; if the barrier is only a few $k_{\\mathrm{B}}T$ at 150 K, then the arrest is a finite-time artifact. On the experimental side, anneal a MAPbI$_3$ single crystal just below the $\\beta$–$\\gamma$ boundary for hours while monitoring neutron or X-ray diffuse scattering and molecular orientation; appearance of the $\\gamma$-phase orientation pattern would refute the claim that the transition is effectively forbidden in accessible conditions.","supporting_citations":[{"cited_title":"P.; Simm, G.; Ortner, C.; Csányi, G","cited_arxiv_id":null,"evidence_quote":"Supplies the equivariant message-passing neural-network architecture used for the production-scale molecular dynamics."},{"cited_title":"J.; Li, Z.; Csányi, G.; Walsh, A","cited_arxiv_id":null,"evidence_quote":"Supplies the tilt-correlation polarity, correlation-length, and molecular-orientation descriptors used for phase classification."},{"cited_title":"Dynamic Nanodomains Dictate Macroscopic Properties in Lead Halide Perovskites","cited_arxiv_id":"2404.14598","evidence_quote":"Provides the experimental diffuse-scattering and neutron data identifying an Im-3 a+a+a+ phase that the low-temperature FAPbX3 assignment is checked against."},{"cited_title":"On-the-fly active learning of in- teratomic potentials for large-scale atomistic simulations.The Journal of Physical Chemistry 35 Letters 2020, 11, 6946–6955","cited_arxiv_id":null,"evidence_quote":"Describes the on-the-fly Bayesian-error data collection strategy used to build the training sets."},{"cited_title":"Phase Transition Kinetics of MAPbI 3 for Tetragonal-to- Orthorhombic Evolution.JACS Au 2023, 3, 1205–1212","cited_arxiv_id":null,"evidence_quote":"Gives DFT kinetics for the tetragonal-to-orthorhombic transformation of MAPbI3 that support the MA-rearrangement barrier interpretation."},{"cited_title":"M.; McMahon, A","cited_arxiv_id":null,"evidence_quote":"Provides neutron-scattering measurements of methylammonium reorientation dynamics consistent with the required molecular rearrangement."},{"cited_title":"H.; Siaw, T","cited_arxiv_id":null,"evidence_quote":"Documents the universal molecular reorientation behavior in hybrid lead iodide perovskites that underlies the FA orientation continuity."},{"cited_title":"A.; Ceder, G","cited_arxiv_id":null,"evidence_quote":"Documents the systematic softening of universal machine-learning potentials that the authors invoke to explain their lowered transition temperatures."},{"cited_title":"Octahedral Tilt-Driven Phase Transitions in BaZrS3 Chalcogenide Perovskite","cited_arxiv_id":null,"evidence_quote":"Provides the group-theoretical analysis of octahedral tilt transitions that underpins the free/guided/limited pathway classification."}],"review_version":1}