{"id":"c3d959a8-00d0-4c42-9f57-4d6785c79b39","arxiv_id":"2505.08732","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"New machine learning potentials for beta-tin reproduce the experimentally observed structure and growth morphology of deformation twins, showing that twin growth proceeds via double-layer twinning disconnections and low-energy PA/AP facets.","lead":"This paper develops two machine learning interatomic potentials for tin and uses them to simulate how twin boundaries move in beta-tin, the material that emits the familiar 'cry of tin' when bent. The simulations match electron microscope images of real twins and reveal faceted boundary structures made of moving steps and a previously unrecognized low-energy interface type.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Missing DFT validation of twin-boundary energetics leaves the h(2) disconnection mechanism and AP/PA facet claims resting on unverified ML-potential transferability.","rationale":"The reader's weakest assumption is exactly the transferability of the ML potentials to twin-boundary and disconnection environments, with the failed DFT validation as the key gap. My stress-test identifies the same load-bearing concern and does not find a separate issue that would change the verdict. The paper has real strengths: the potentials and training data are released, the bicrystallographic analysis is systematic, and the qualitative TEM comparison provides some support. But the central mechanistic claims, especially the h(2) disconnection-mediated migration and the low-energy AP/PA facet interpretation, depend on an energy landscape that has not been directly checked against DFT for twin boundaries. Because the authors themselves attempted the check and could not complete it, and because the experimental evidence cannot resolve individual disconnections or directly measure coupling factors, the conditional verdict is appropriate. A focused DFT calculation of the CTB energies and gamma-surface would settle whether the concern lands; until then, the manuscript should remain conditional rather than being accepted as fully verified.","tokens_in":23876,"tokens_out":6146,"duration_ms":68515,"concrete_test":"Perform a converged DFT calculation of the (301) and (101) CTB energies, for example using smaller repeat cells, Methfessel-Paxton smearing, a denser k-mesh, and the ML-relaxed structures as starting points, and compare the values and ordering with Table 1. Additionally compute the CTB gamma-surface along the h(2) disconnection Burgers-vector direction. If the DFT CTB energy ordering or gamma-surface minima deviate substantially from the MTP/RANN predictions, the h(2) migration mechanism and AP/PA facet stability are not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the MTP and RANN potentials reveal the true twin-boundary structure, migration mechanisms, and facet energetics in beta-Sn. This requires these potentials to be accurate in twin-boundary and disconnection environments, but they were trained on bulk, surface, defect, and perturbed configurations, not on explicit twin boundaries. The only direct ab initio check attempted, the DFT calculation of the (301) and (101) CTB energies, did not converge (Sec. 3.1), and the supplementary stacking-fault comparison (Fig. S1) does not constrain twin-boundary or disconnection energetics. The TEM comparison is also qualitative and involves a single boundary: individual h(2) disconnections are not atomically resolved, the shear coupling factor has never been measured directly, and the K2 loading state used to reproduce the experimental growth morphology is inferred rather than measured. Agreement between the two ML potentials, and with MEAM on some features, is encouraging but does not establish accuracy. A further symptom of the same gap is the c/a mismatch: MTP and RANN give c/a = 0.538 and 0.540 versus the experimental 0.546, which changes the geometric twinning shear from about 0.097 to about 0.12 at 0 K. If converged DFT shows materially different CTB energies or a different CTB gamma-surface, the predicted h(2) mechanism and low-energy AP/PA facet picture could be potential artifacts rather than properties of beta-Sn.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops two machine-learning interatomic potentials (an MTP and a RANN potential) for β-Sn, trains them on a DFT database spanning multiple phases, defects, and perturbed configurations, and uses them to study the structure, energy, and shear-coupled migration of (301) and (101) coherent twin boundaries (CTBs). The simulations predict that both CTBs migrate by double-layer h(2) disconnections with small negative shear coupling factors around -0.1, and that low-energy asymmetric PA/AP facets form during twin growth. These predictions are compared against high-resolution TEM observations of a deformed (101)-oriented single crystal, where a semi-coherent near-(101) growth facet is observed and reproduced in simulation under an inferred K2 loading state. The authors also perform bicrystallographic analysis with oiLAB and lattice-matching calculations, and compare the ML potentials against a MEAM potential. The potentials, training data, and sample inputs are made publicly available.","tokens_in":24173,"tokens_out":5110,"duration_ms":54158,"significance":"If the central claims hold, this is the first direct atomistic characterization of twin-boundary migration in a tetragonal metal, and it provides a mechanistic explanation for the experimentally observed faceted twin morphologies in β-Sn. The paper's strengths include: (1) two independently trained ML potentials that agree on the main twinning features, (2) a publicly available training database and potentials, which directly supports reproducibility, (3) a direct comparison to HRTEM and diffraction data, including a quantitative step-height/spacing analysis, (4) a bicrystallographic framework that ties the observed facets and disconnections to the dichromatic pattern, and (5) an explicit discussion of several remaining limitations. The discovery of PA/AP facets as low-energy interfaces analogous to PB/BP facets in HCP metals is an interesting and potentially transferable concept.","major_comments":[{"comment":"The DFT validation of the (301) and (101) CTB energies is reported as 'unable to converge' with no further details on the attempted calculation parameters (e.g., k-point sampling, smearing, or relaxation protocol). This missing support is load-bearing because the training set described in Sec. S.1 includes bulk, surfaces, point defects, and perturbed configurations, but not explicit twin boundaries or disconnection environments. The only ab initio check provided, the stacking-fault comparison in Fig. S1, does not directly constrain twin-boundary or disconnection energetics. The authors should either supply converged DFT reference energies for at least the two CTBs (perhaps using smaller supercells or a different electronic-structure approach), or alternatively provide a sensitivity analysis showing how the predicted energy ordering, h(2) disconnection stability, and AP/PA facet energies vary under plausible changes to the potential. Without this, the quantitative predictions in Table 1 and the mechanism in Fig. 6 remain unverified in the very environments they are meant to describe.","section":"Sec. 3.1 and Sec. S.1"},{"comment":"The c/a ratios of the MTP and RANN potentials are 0.538 and 0.540, respectively, versus the experimental 0.546, and the geometric twinning shear in Fig. 4 consequently increases from the experimental 0.097 to 0.116-0.122 at 0 K. While the 300 K coupling factors are close to the experimental value, the 0 K values are about 20% more negative. The statement 'less than 5% variation between theory and simulation for 0 K data' compares the simulations to the geometric formula, not to the experimental value; the experimentally relevant comparison is to the measured or inferred value of about -0.1. The authors should quantify the sensitivity of the predicted coupling factor to c/a and to the underestimated C44 elastic constant (Table S1), and explicitly discuss why the finite-temperature values, rather than the 0 K values, are the appropriate comparison to the experimental coupling factor. This would clarify whether the residual discrepancy reflects the potentials' lattice-parameter error or a genuine mechanistic difference.","section":"Sec. 3.2, Table 1, and Fig. 4"},{"comment":"The experimental validation relies on a single twin boundary, and the loading state (K2 shear) used to reproduce the observed semi-coherent near-(101) growth morphology is inferred from the morphology itself rather than measured. In addition, the HRTEM images do not atomically resolve individual h(2) disconnections; the comparison is based on an estimated total step height and an average defect spacing. As written, the paper presents the h(2) disconnection mechanism and the coupling-factor sign/magnitude as simulation predictions that are 'consistent with' the experiments, but it should be stated more prominently that the experiments do not directly confirm the atomic-scale mechanism. The authors could strengthen this discussion by suggesting concrete experimental methods (e.g., atomic-resolution STEM or in-situ straining experiments with diffraction contrast) that could directly image the h(2) disconnections or measure the coupling factor, and by noting which of the simulated features are robust predictions versus inferred interpretations.","section":"Sec. 3.3 and Fig. 7"},{"comment":"The discovery claim for PA/AP facets as low-energy interfaces important to twin growth in β-Sn is based on simulation-predicted facet structures and on angular matching to prior TEM micrographs that do not have atomic resolution. The authors themselves note in Sec. 4 that 'atomic resolution in the TEM data would be needed' and that the character of AP/PA interfaces 'has yet to be directly confirmed with atomic resolution.' Because this is one of the three headline contributions stated in the Introduction and Conclusion, the claim should be explicitly qualified in the abstract and conclusions, or additional evidence should be provided—for example, image simulations of the proposed AP/PA structure compared to the experimental micrographs, or a quantitative analysis of the facet angle distribution. As it stands, the tone of the abstract ('a discovery of low energy asymmetric PA/AP interfaces important to twin growth') is stronger than the current evidence supports.","section":"Sec. 4 and Conclusions"}],"minor_comments":[{"comment":"The heading 'Addition atomistic data' should read 'Additional atomistic data'.","section":"SI, Sec. S.3"},{"comment":"The symbol γ is used both for the c/a ratio in Eq. (3) and for interface energy throughout the text and Table 1; this notational clash could confuse readers. Consider using a different symbol for the axial ratio, such as η or ζ.","section":"Sec. 3.2, Eq. (3)"},{"comment":"The red angular marker 'calibrated to MTP simulation data' is not described with a numerical uncertainty; adding the MTP-predicted angle and the experimental measurement uncertainty on the marker itself would make the 2° deviation easier to interpret.","section":"Sec. 3.3, Fig. 7a"},{"comment":"The experimental coupling-factor entries mix sources with different sign conventions (e.g., -0.0982 [54], |β| = 0.0978 [2], 0.113 [3]). Citing all three without a unified sign convention may confuse readers; a short footnote explaining the convention would help.","section":"Table 1"},{"comment":"The text says 'the (1 0 1) twin boundary exhibits a larger yield strain and flow stress for migration than the (3 0 1) twin boundary by at least an order of magnitude', but the yield strains of 0.01 versus 0.001 are different by exactly an order of magnitude and the flow stresses differ by more; 'by at least an order of magnitude' is slightly imprecise.","section":"Sec. 3.3, p. 18"}],"recommendation":"major_revision","confidential_remarks":"The central scientific contribution is timely and the experimental/simulation comparison is a strong point, but the missing DFT convergence is a genuine technical risk that the authors should address. I would not reject on that basis, but I would urge the editor to require an attempt at converged DFT reference energies for at least the (301) and (101) CTBs, even in reduced supercells, or a well-argued sensitivity analysis. The paper's publication would be substantially strengthened if the abstract and conclusions were rebalanced to reflect the current limitations on AP/PA facet confirmation and on the inference of the loading state. The manuscript is a good fit for a materials-science journal of the Acta Materialia type."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my read on 2505.08732. It is worth taking seriously: it is the first direct atomistic simulation of twin boundary migration in a tetragonal metal, and it comes with two new ML potentials (MTP and RANN) plus open training data and a direct TEM comparison. The central picture—(301) and (101) CTBs move by h(2) disconnections with small negative coupling factors around -0.1, and low-energy asymmetric PA/AP facets are important for twin growth—is coherent and well-argued.\n\nWhat is genuinely new: the atomistic characterization of twin structure and migration in beta-Sn, the discovery of PA/AP facets as a distinct interface type (analogous to PB/BP in HCP), and a bicrystallographic framework that explains the observed semi-coherent near-(101) growth facet as a consequence of K2 loading. The experimental HRTEM of the (301) twin, with measured misorientation and step spacing, provides a real anchor that the simulations are not just fitting noise. The paper is also honest about the MEAM potential's failure, which is a useful control.\n\nThe soft spots are real but not fatal. The biggest is that the twin boundary energies and the gamma-surface that controls the h(2) mechanism are entirely from ML potentials that were not trained on twin boundary environments. The DFT validation stated in Sec. 3.1 did not converge, so there is no ab initio check on the very quantities the story depends on. Stacking fault comparisons (Fig. S1) are a good sanity check but do not constrain CTB or disconnection energetics. The c/a mismatch matters too: 0.538–0.540 versus experimental 0.546 shifts the ideal twinning shear from 0.097 to about 0.12, a ~25% relative change. That is more than the \"less than 5%\" phrasing in the text suggests. Second, the identification of PA/AP facets in experiments is indirect—the authors say atomic resolution is needed. Third, the loading state used to reproduce the experimental growth morphology (K2 shear) is inferred, not measured.\n\nNone of this kills the paper. The two ML potentials agree on the key mechanisms, and the TEM comparison for the (301) twin is persuasive. The right fix is to add some DFT reference data for the CTB energies and, ideally, the disconnection core, using smaller supercells or better convergence strategies. The authors should also address the c/a mismatch explicitly rather than subsuming it into the formula agreement.\n\nWho is this for: anyone working on twinning in low-symmetry metals, ML interatomic potentials, or beta-Sn reliability. It deserves a serious referee. I would accept it for review, with the expectation that the DFT gap gets addressed in revision.","headline":"First atomistic study of twin migration in beta-Sn, with two solid ML potentials and a real TEM anchor; the missing DFT validation of twin-boundary energetics is a genuine but addressable weakness.","tokens_in":24749,"tokens_out":3579,"would_cite":true,"duration_ms":36185,"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":"Machine-learning interatomic potentials that reproduce electron micrographs of deformation twins in β-Sn reveal that both (301) and (101) twin boundaries migrate by small double-layer disconnection steps, and expose low-energy asymmetric…","keywords":["twin boundaries","beta-Sn","tetragonal crystal","machine learning interatomic potentials","moment tensor potential","neural network potential","disconnections","shear coupling"],"falsifier":"A converged density-functional-theory calculation of the (301) and (101) coherent twin boundary energies in β-Sn would settle the potential transferability question: the MTP and RANN potentials predict these energies near 57–79 mJ m$^{-2}$ with (301) lower, while MEAM predicts (101) lower, and the paper reports that its DFT attempts did not converge. An independent DFT result that places either energy outside this window, or reverses the ordering, would indicate the ML potentials' twin-boundary description is wrong.","tokens_in":23694,"feed_emoji":"🔬","tokens_out":10941,"duration_ms":103760,"temperature":0.7,"pith_summary":"This paper aims to establish that machine-learning interatomic potentials can reproduce the atomic structure and migration of twin boundaries in β-Sn, a low-symmetry tetragonal metal whose deformation twinning produces tin's 'cry' and contributes to solder-joint failure. By fitting a moment tensor potential and a rapid neural network potential, then comparing direct atomistic simulations with high-resolution transmission electron microscopy of deformed tin, the authors show that both the (301) and (101) coherent twin boundaries migrate by the nucleation and glide of $h^{(2)}$ disconnections — steps two interatomic planes high — with small shear coupling factors near $-0.1$. The simulations also expose low-energy asymmetric PA/AP facets that act as growth facets during twin expansion, analogous to prismatic-basal facets in hexagonal metals. If the claims hold, this is the first direct atomistic description of twin boundary migration in a tetragonal metal, and the newly released potentials give other researchers a tool for modeling twinning and phase transformations in tin.","feed_headline":"Tin twins migrate via tiny disconnection steps","feed_subtitle":"New machine-learned potentials match TEM images and explain tin's cry and solder embrittlement.","key_machinery":"The load-bearing mechanism is the twinning disconnection, a line defect that combines a step in the twin boundary with a Burgers vector, coupling boundary motion to macroscopic shear; the specific object is the $h^{(2)}$ disconnection, a step two interatomic planes tall with a small Burgers vector that produces the measured coupling factor near $-0.1$. The companion machinery is bicrystallography: the $\\Sigma_2$ dichromatic pattern of the β-Sn lattice, built from two interpenetrating body-centered tetragonal lattices, places the (301) and (101) coherent twin boundaries exactly $90^\\circ$ apart when the axial ratio has its ideal value, and predicts zero twinning shear. The non-ideal $c/a$ ratio breaks that symmetry, generating the observed misorientation angles, the finite Burgers vectors of the twinning disconnections, and the disclination character of asymmetric facets. That geometric framework, together with the energies computed from the new potentials, explains why PA/AP facets are stable low-energy features and why they appear at specific angles in twin microstructures.","core_discovery":"The central claim is that twin boundaries in tetragonal β-Sn migrate by the nucleation and glide of $h^{(2)}$ twinning disconnections: steps two interatomic planes high carrying a small Burgers vector, which yields a negative shear coupling factor of about $-0.1$ for both the (301) and the (101) coherent twin boundary. The paper further claims that twin growth in β-Sn proceeds through low-energy asymmetric facets, called PA/AP interfaces, in which a prismatic $\\{101\\}$ plane of one grain joins an A-type $\\{100\\}$ plane of the other, and that these facets accumulate and release twinning disconnections, giving twin growth flexibility. The support comes from comparing newly fitted machine-learning potentials (MTP and RANN) with high-resolution TEM of deformation twins: the simulated structures, including a semi-coherent near-(101) growth interface, match the micrographs, whereas the earlier MEAM potential does not reproduce the (101) migration mode. If correct, this resolves the longstanding question of how β-Sn twins grow atom by atom and provides interface energies, coupling factors, and critical stresses for use in mesoscale deformation models.","pith_inferences":["We infer that the near-zero shear coupling factor, combined with the extremely low flow stress for (301) migration (about 0–5 MPa with the MTP), implies that twin boundary motion should be an easy deformation channel in β-Sn at low temperatures; this follows from the paper's numbers but is not a claim the paper itself tests.","Because the ML potentials were trained without any explicit twin boundary or disconnection structures, the paper's success suggests that the same training recipe may transfer to other low-symmetry metals; a direct test would be to fit an MTP for another body-centered tetragonal material and check whether its twin boundaries migrate through the same $h^{(2)}$ mechanism.","The paper's observation that long AP facets in prior experiments (around 50 nm) are an order of magnitude longer than simulated facets hints that a different accommodation mechanism, perhaps misfit dislocation networks, operates at larger scales; the authors note this as open, and our inference is that testing it requires simulations with box sizes of hundreds of nanometers."],"forward_implications":["The (301) and (101) twin boundaries should be treated as sharing the same $h^{(2)}$-mediated migration mechanism with a common shear coupling factor near $-0.1$, as earlier inferred from diffraction geometry and now directly observed in simulations.","Mesoscale models of twinning in β-Sn now have quantitative inputs: twin boundary energies around 57–79 mJ m$^{-2}$, AP facet energies near 55–88 mJ m$^{-2}$, and critical flow stresses ranging from a few MPa for (301) up to roughly 50 MPa for (101) with the MTP.","Equilibrium Wulff shapes of (301) and (101) twin inclusions should contain PA/AP facets, so these asymmetric interfaces belong in any geometric model of twin growth.","Simulation studies that use the MEAM potential for β-Sn twinning should be revisited, since MEAM predicts a different mechanism for (101) migration and does not produce the experimentally observed semi-coherent growth interface.","The released MTP and RANN potentials are suitable for modeling other extended-defect phenomena in tin, including phase transformations, dislocation-grain boundary interactions, and detwinning."],"supporting_citations":[{"why":"Provides experimental observation of twinning in tin lamellae and the measured twinning shear that the simulations are compared against.","marker":"[2]"},{"why":"Supplies the c/a-dependent twinning shear formula and the shuffle analysis used to interpret the h(2) disconnection motion.","marker":"[3]"},{"why":"Defines the disconnection framework and the h(n) step notation used to identify the migration mechanism.","marker":"[20]"},{"why":"Provides the MEAM potential that is the classical baseline compared against the ML potentials.","marker":"[33]"},{"why":"Gives the predecessor neural network potential and the DFT stacking fault energies used to argue transferability.","marker":"[36]"},{"why":"Supplies the Moment Tensor Potential formalism on which one of the new potentials is built.","marker":"[42]"},{"why":"Introduces the prismatic/basal interface analogy in HCP crystals that motivates the PA/AP facet discovery.","marker":"[64]"},{"why":"Provides the bicrystallographic tool used to enumerate facet inclinations and disconnection modes for β-Sn twins.","marker":"[63]"}],"fun_headline_variants":["Tin twin growth explained by tiny disconnection steps","Machine-learned potentials match tin twin TEM images","Beta-Sn twins migrate via 2-plane disconnection steps","Low-energy facets govern beta-Sn twin growth","ML potentials unlock beta-Sn twin migration secrets"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The machine-learning potentials must be accurate for twin-boundary and disconnection environments even though no explicit twin or grain boundary structures appear in their training data; if the potentials represent the twin boundary energy landscape incorrectly, the predicted facet stabilities, energies, and migration mechanisms would all be unreliable.","fun_headline_variants_meta":{"raw":{"variants":["Tin twin growth explained by tiny disconnection steps","Machine-learned potentials match tin twin TEM images","Beta-Sn twins migrate via 2-plane disconnection steps","Low-energy facets govern beta-Sn twin growth","ML potentials unlock beta-Sn twin migration secrets"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000172,"raw_usage":{"total_tokens":1316,"prompt_tokens":1028,"completion_tokens":288,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":644,"completion_tokens_details":{"reasoning_tokens":214}},"tokens_in":644,"tokens_out":288,"duration_ms":2975,"temperature":1.0,"reasoning_tokens":214,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:47:38.284768+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A converged density-functional-theory calculation of the (301) and (101) coherent twin boundary energies in β-Sn would settle the potential transferability question: the MTP and RANN potentials predict these energies near 57–79 mJ m$^{-2}$ with (301) lower, while MEAM predicts (101) lower, and the paper reports that its DFT attempts did not converge. An independent DFT result that places either energy outside this window, or reverses the ordering, would indicate the ML potentials' twin-boundary description is wrong.","supporting_citations":[{"cited_title":"Direct observation of twinning in tin lamellae","cited_arxiv_id":null,"evidence_quote":"Provides experimental observation of twinning in tin lamellae and the measured twinning shear that the simulations are compared against."},{"cited_title":"Christian, The Theory of Transformations in Metals and Alloys , (Newnes, 2002)","cited_arxiv_id":null,"evidence_quote":"Supplies the c/a-dependent twinning shear formula and the shuffle analysis used to interpret the h(2) disconnection motion."},{"cited_title":"Disconnections and other defects associated with twin interfaces","cited_arxiv_id":null,"evidence_quote":"Defines the disconnection framework and the h(n) step notation used to identify the migration mechanism."},{"cited_title":"Atomistic simulations of pure tin based on a new modified embedded-atom method interatomic potential","cited_arxiv_id":null,"evidence_quote":"Provides the MEAM potential that is the classical baseline compared against the ML potentials."},{"cited_title":"Hybrid interatomic potential for Sn","cited_arxiv_id":null,"evidence_quote":"Gives the predecessor neural network potential and the DFT stacking fault energies used to argue transferability."},{"cited_title":"On the importance of prismatic/basal interfaces in the growth of (-1012) twins in hexagonal close-packed crystals","cited_arxiv_id":"1301.1902","evidence_quote":"Introduces the prismatic/basal interface analogy in HCP crystals that motivates the PA/AP facet discovery."},{"cited_title":"Interface dislocations and grain boundary discon- nections using smith normal bicrystallography","cited_arxiv_id":null,"evidence_quote":"Provides the bicrystallographic tool used to enumerate facet inclinations and disconnection modes for β-Sn twins."}],"review_version":1}