{"id":"cfb16046-451d-4e1d-b226-f769d10e4a85","arxiv_id":"2607.24107","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A new SSCHA-based phase-field method with fractional site occupancy computes site-resolved chemical potentials and predicts Ag segregation at Cu grain boundaries.","lead":"The authors extend a stochastic self-consistent harmonic approximation (SSCHA) free-energy framework to allow fractional site occupancy, making chemical potentials differentiable and enabling atomic-scale phase-field simulations of dopant distributions. It reproduces Ag distributions and lattice expansion in bulk Cu and predicts Ag segregation sites at a Cu grain boundary.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Predicted segregation-site ranking is not demonstrated to be independent of the arbitrary kinetic coefficient Mij and simulation duration; it may be a kinetic artifact.","rationale":"The reader's conditional verdict is appropriate, but the strongest unresolved issue is not the Gaussian approximation per se — that is partially validated by the bulk lattice-constant agreement with MD — but rather the lack of evidence that the grain-boundary simulations have reached a physically meaningful state. The arbitrary mobility Mij directly enters the dynamics that produce the predicted site preferences, and the cooling protocol is explicitly non-equilibrium (starting from 1200 K and freezing at 300 K). Since the central claim includes these segregation predictions, they need sensitivity and convergence checks. A direct equilibrium calculation of the site occupations would also be decisive. Our proposed test would settle whether the pattern is robust to the unknown kinetic parameter and simulation length.","tokens_in":11842,"tokens_out":12761,"duration_ms":116255,"concrete_test":"Run the same Σ5(310) grain-boundary simulation for NAg = 4, 8, and 16 with Mij = 0.001, 0.01, 0.1, 1, and 10 eV⁻¹s⁻¹ (or equivalently, scale the time step by factors of 10⁻³, 10⁻², 1, 10, 10²) and extend each run until the maximum change in any ci over 10⁴ steps is below 1e-3. If the ranked order of site occupancies and the final values of ci at each site are unchanged across this range, the kinetic-coefficient concern is resolved. If the ranking or values shift, the paper's segregation-site prediction is not robust and must be reformulated as an equilibrium calculation (e.g., solving µ_i = µ subject to fixed total N) rather than a kinetic outcome.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central prediction — that Ag segregates first to vertex, then near-vertex, then bottom sites of the triangular motif in the Cu Σ5(310)[001] grain boundary (Fig. 5) — is the output of the occupation dynamics Eq. (11) with an arbitrarily chosen kinetic coefficient Mij = 0.1 eV⁻¹s⁻¹ (the authors state this mobility is unknown), combined with a non-equilibrium cooling protocol that starts from 1200 K distributions and freezes them at 300 K. The paper reports no convergence criterion for the occupation probabilities; Figs. 4 and 5 show only a few hundred steps. At equilibrium, the distribution would satisfy µ_i = µ for all i and would be independent of Mij, but no evidence is provided that the simulations reached that state. Because Mij sets the physical time scale, the 'frozen-in' pattern and the ordering of sites (vertex → near-vertex → bottom) could depend sensitively on this parameter and on the number of steps. This directly threatens the predictive claim of the paper, which is not merely about the formal methodology but about a specific segregation pattern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an atomic-scale phase-field method that combines stochastic self-consistent harmonic approximation (SSCHA) with fractional site occupancies. A variational free energy is constructed from the Gibbs–Bogoliubov inequality (Eqs. 6–8), with Gaussian vibrational clouds and continuous occupation variables c_i. From this functional the authors derive site-resolved chemical potentials (Eq. 19), atomic equations of motion (Eqs. 10, 17), and a pressure/volume-control scheme (Eqs. 20–23). The method is tested on Ag-doped bulk Cu, where the computed temperature-dependent lattice constants are compared with MD results (Fig. 3), and then applied to Ag segregation at a Cu Σ5(310)[001] grain boundary, where it predicts preferential occupation of vertex, near-vertex, and bottom sites of the triangular motif (Figs. 4–5). The manuscript claims this extends phase-field modeling to atomic-scale dopant segregation.","tokens_in":12105,"tokens_out":8965,"duration_ms":91523,"significance":"If the method is robust, it is a useful extension of SSCHA toward atomistic phase-field modeling with many-body interatomic potentials. The derivation from the Gibbs–Bogoliubov inequality is standard and the lattice-constant check against MD provides a meaningful external benchmark. The use of a machine-learned potential (MACE) and GPU-accelerated Sobol Monte Carlo sampling is a practical strength. However, the central segregation-site prediction currently rests on an arbitrarily chosen kinetic coefficient and on finite-time occupation evolution without demonstrated convergence or equilibrium. The paper's claim to identify primary, secondary, and tertiary segregation sites is therefore not yet supported. The limitations stated by the authors—especially the neglect of vibrational correlations and of complexion transitions—are important and should be more prominently connected to the scope of the claims.","major_comments":[{"comment":"The central segregation ranking is produced by the occupation dynamics Eq. (11) with Mij = 0.1 eV⁻¹ s⁻¹, which the authors state is unknown. No convergence criterion is reported for c_i, and Figs. 4–5 show only finite minimization steps. In equilibrium, the ranking should be independent of Mij and of runtime, but no evidence is given that the plotted states are converged. Please provide a convergence check (e.g., max_i |dc_i/dt| or the spread of µ_i versus step) and an Mij-sensitivity test. If the states are deliberately non-equilibrium quenches, state this explicitly and justify why the ordering is robust.","section":"§Governing equations and Simulation details, Eq. (11), Fig. 5"},{"comment":"The chemical potentials are evaluated by Monte Carlo with a sample size of only 100 per site, and no statistical error bars are reported for µ_i or for the lattice constants in Fig. 3. If the site-to-site differences that determine the vertex→near-vertex→bottom ranking are smaller than the MC noise, the ranking is not statistically supported. Report standard errors for the relevant quantities and increase Nsample if necessary.","section":"Simulation details, Eq. (19), Fig. 3"},{"comment":"The atomic mobility κ is taken from Cu²⁺ in water at 298 K and the volume mobility M′ is set to 10.0 Pa⁻¹ s⁻¹ without physical calibration or sensitivity analysis. These parameters control the evolution of atomic positions and volume, which are coupled to the occupancy dynamics through the free energy. Demonstrate that the final segregation patterns are insensitive to κ and M′, or calibrate them; otherwise the structural contribution to the predicted pattern is uncontrolled.","section":"Simulation details, Eq. (17) and volume control"},{"comment":"The 300 K segregation patterns are initialized from 1200 K distributions and ‘frozen in’ during cooling, but no cooling protocol or quench schedule is specified. It is therefore unclear whether Fig. 5 represents equilibrium states at 300 K or kinetic artifacts of the cooling procedure. In addition, the authors' own limitation (1) excludes complexion transitions, which are relevant to grain-boundary segregation. Please clarify the thermodynamic status of the plotted distributions and temper the predictive claim accordingly.","section":"§Results, Fig. 5, Limitations"},{"comment":"In Eq. (8), a single covariance Σ_i is used for both host and dopant species at a partially occupied site: only the thermal de Broglie wavelength differs in the vibrational term, and the interaction term samples a common Gaussian cloud regardless of the sampled species. This imposes identical vibrational amplitudes on Cu and Ag, which may bias the site-resolved chemical potentials, especially for a size-mismatched dopant. Please discuss this approximation or test its effect by allowing species-dependent covariance matrices.","section":"§Methods, Eq. (8)"}],"minor_comments":[{"comment":"The text says 'The calculation of the chemical potential, Eq. (18)' but the chemical potential is defined in Eq. (19). Please correct the cross-reference.","section":"Limitations section"},{"comment":"The abstract states that the method 'successfully reproducing Ag distributions in bulk Cu,' but Fig. 2 shows a finite-size-stabilized ordered-like state in a system where Cu–Ag is experimentally prone to phase separation. Consider softening the wording to 'explores low-energy occupancy patterns' rather than 'reproducing distributions.'","section":"Abstract and Fig. 2"},{"comment":"The color scale is described as the same in all three figures, but no color bar is shown. Since the quantitative occupation probabilities are central to the ranking, a color bar with numerical values would improve interpretability.","section":"Figs. 2, 4, 5"},{"comment":"The caption says 'shown at (a) 0, (b) 10, and (c) 100 minimization steps,' while the main text refers to these as 'Figs. 4(b) and 4(c)' without specifying the step numbers in the same notation. This is a minor inconsistency in presentation that should be harmonized.","section":"Fig. 4 caption"},{"comment":"The use of Sobol sequences and Nsample=1000 for the free energy is a good practical choice. It would strengthen the paper to state whether the same low-discrepancy sequence is used for the chemical-potential estimate (Nsample=100) and to report the reproducibility across different sequence lengths.","section":"MC sampling"}],"recommendation":"major_revision","confidential_remarks":"The core derivation is sound and the validation against MD lattice constants is a positive check, but the central segregation-site prediction is not yet robust because it depends on an arbitrary kinetic coefficient and on finite-time dynamics without convergence evidence. The requested additions—convergence checks, Mij sensitivity, MC error bars, and clarification of the cooling protocol—are within the scope of a revision. If those cannot be provided, the paper should be reframed as a methodological demonstration rather than a validated prediction of dopant segregation patterns. The novelty relative to Refs. 34–36 is incremental but appears sufficient for a methods-oriented paper if the above points are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Core take: the free-energy functional is a genuine step forward, but the segregation hierarchy in Fig. 5 is not yet demonstrated to be kinetically robust; treat it as provisional.\n\nWhat's new and good: extending SSCHA to fractional site occupancy with many-body potentials and anisotropic Gaussians, and deriving a site-resolved chemical potential as an explicit free-energy derivative, is a real contribution beyond Li et al.'s pair-potential, isotropic limit. The bulk lattice-constant agreement with MD is a healthy external check on the free-energy functional, even if the test is a small 32-atom cell. The paper is also honest about its limitations.\n\nSoft spots: the grain-boundary segregation prediction depends on the occupancy exchange mobility Mij, which is unknown and set to 0.1 eV^-1 s^-1, and the atomic mobility is taken from Cu^2+ in water. Equilibrium occupancy should be M-independent, but nothing demonstrates that the 300 K simulations reached equilibrium; they are started from 1200 K patterns and frozen in. With only a few hundred steps and no reported convergence criterion, the vertex → near-vertex → bottom ordering could be an artifact of the quench schedule and arbitrary kinetics. The bulk validation supports the free energy, but it does not validate the kinetic part. Also, no error bars are given on Monte Carlo estimates (1000 samples for F, 100 for µ), and there is no code/data release, which hampers reproduction. The finite-size-stabilized 'ordered-like' bulk state is acknowledged as not the true thermodynamic phase.\n\nVerdict: the central derivation looks sound, and the paper's own limitations are stated. The method deserves a serious referee. For acceptance, I would want a convergence analysis of the occupancy dynamics, a check that the site ordering is insensitive to Mij and cooling rate, and ideally error bars and code release. This paper is for people working on atomistic thermodynamics, segregation, and SSCHA-based methods; they will get value from the formalism even if they do not yet trust the specific grain-boundary ranking.\n\nRecommendation: send to peer review, with the kinetic-sensitivity issue as the central requested revision.","headline":"Solid methodological extension of SSCHA to fractional site occupancy, but the grain-boundary site-ordering prediction rests on unvalidated kinetics; worth peer review with convergence and sensitivity checks.","tokens_in":12550,"tokens_out":1987,"would_cite":true,"duration_ms":19115,"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":"By treating site occupation as a continuous variable inside a variational free-energy functional, this paper derives site-resolved chemical potentials and predicts where Ag dopants segregate at a Cu grain boundary.","keywords":["phase-field modeling","atomic-scale simulation","stochastic self-consistent harmonic approximation","fractional site occupancy","site-resolved chemical potential","dopant segregation","grain boundary","Cu–Ag alloy"],"falsifier":"Compute the site-resolved segregation free energy of Ag at the Σ5(310)[001] Cu grain boundary by direct thermodynamic integration, or measure the 300 K site occupancy by atom-probe tomography: if the order is not vertices first, then near-vertex sites, then bottom sites, the chemical-potential functional is not capturing equilibrium segregation.","tokens_in":11726,"feed_emoji":"⚛️","tokens_out":7010,"duration_ms":60205,"temperature":0.7,"pith_summary":"Phase-field modeling has traditionally been a continuum-scale tool; this paper tries to bring it down to the atomic scale. The key move is to let each lattice site carry a continuous occupation probability inside a variational free energy built from the stochastic self-consistent harmonic approximation, so the free energy becomes differentiable with respect to site occupation. That differentiability yields a site-resolved chemical potential, which supplies the thermodynamic driving force for atomic-scale pattern formation such as dopant segregation. The authors show the approach reproduces the temperature-dependent lattice constant of Ag-doped bulk Cu and predicts a specific segregation hierarchy at a Cu Σ5(310)[001] grain boundary: Ag atoms fill the vertices of the triangular structural motif first, then near-vertex sites, then bottom sites. A sympathetic reader would care because this is a route to predicting, not just rationalizing, where dopants sit in real microstructures.","feed_headline":"Ag dopants pick grain-boundary vertices first, atomic model shows","feed_subtitle":"Continuous site-occupancy free energy predicts the order of Ag segregation sites at a Cu grain boundary.","key_machinery":"The central object is the variational free-energy functional in which each atom is represented by a general anisotropic Gaussian probability cloud and each site carries a continuous occupation probability. The functional derivative of the free energy with respect to site occupation gives the site-resolved chemical potential, and the negative gradient with respect to atomic position gives the mean force that moves the Gaussian clouds. Because the free energy is evaluated by stochastic sampling of a many-body interatomic potential, the machinery converts the discrete question of which site is occupied by a dopant into a differentiable thermodynamic landscape that can be minimized and evolved i","core_discovery":"The central claim is that a stochastic self-consistent harmonic approximation free-energy functional, extended to fractional site occupancy, is a working atomic-scale phase-field model. Because each site's occupancy is continuous, the free energy is differentiable with respect to occupation, and its derivative gives a site-resolved chemical potential. Using this potential together with the mean force on each Gaussian atomic cloud, the authors evolve both occupation probabilities and atomic positions. They show the model reproduces the temperature-dependent lattice constant of Ag-doped bulk Cu, matching molecular dynamics, and predicts a concrete segregation hierarchy at a Cu Σ5(310)[001] gra","pith_inferences":["Editorial inference: the predicted segregation hierarchy at the Σ5 boundary is the paper's sharpest testable output; a direct site-resolved segregation-energy calculation or high-resolution chemical mapping at 300 K could confirm or refute it independently of the phase-field dynamics.","Editorial inference: the 'ordered-like' Ag occupancy pattern seen in the small bulk Cu–Ag cell is likely a finite-size effect; in a larger cell, phase separation rather than ordering should emerge, and testing that limit would stress the free-energy functional.","Editorial inference: the two kinetic parameters—the occupation-exchange mobility and the atomic mobility taken from an aqueous ion—are placeholders; if the final segregation pattern is sensitive to them, the reported hierarchy would incorporate kinetic bias rather than pure equilibrium thermodynamics.","Editorial inference: the authors' own limitation note points to oxides as the next stress test, since vibrational correlations neglected here become important there; applying the method to an oxide grain boundary would reveal how much the independent-Gaussian approximation limits the approach."],"forward_implications":["For bulk Cu–Ag, the method reproduces the temperature-dependent lattice constant as the dopant number varies, matching molecular-dynamics reference calculations.","For the Σ5(310)[001] Cu grain boundary, the method predicts a clear segregation order—vertices first, then near-vertex sites, then bottom sites—thereby identifying primary, secondary, and tertiary segregation sites.","Because chemical potentials are computed site-resolved, the framework provides a thermodynamic driving force for atomic-scale pattern formation without relying on continuum fields.","The same formalism applies to many-body interatomic potentials and to general anisotropic vibrational shapes, going beyond earlier pair-potential, isotropic-Gaussian treatments.","The framework visualizes atomic-scale pattern formation as probability-density maps, allowing direct comparison with atomic-resolution observations."],"fun_headline_variants":["Atomic-scale phase-field predicts Ag segregation order at Cu grain boundary","Continuous occupancy model reveals Ag's preferred grain-boundary sites","Model pins Ag to vertices first, then bottom sites, at Cu grain boundary","New model predicts Ag dopant hierarchy at Cu grain boundaries","Fractional occupancy phase-field spots Ag's first segregation sites"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise—flagged by the authors in their Limitations section (item 2)—is that independent Gaussian vibrational clouds give a faithful free-energy surface for the Cu–Ag systems studied; if vibrational correlations are significant, the site-resolved chemical potentials and the segregation ranking could change.","fun_headline_variants_meta":{"raw":{"variants":["Atomic-scale phase-field predicts Ag segregation order at Cu grain boundary","Continuous occupancy model reveals Ag's preferred grain-boundary sites","Model pins Ag to vertices first, then bottom sites, at Cu grain boundary","New model predicts Ag dopant hierarchy at Cu grain boundaries","Fractional occupancy phase-field spots Ag's first segregation sites"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001275,"raw_usage":{"total_tokens":5049,"prompt_tokens":737,"completion_tokens":4312,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":4226}},"tokens_in":481,"tokens_out":4312,"duration_ms":26205,"temperature":1.0,"reasoning_tokens":4226,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T23:01:45.838347+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the site-resolved segregation free energy of Ag at the Σ5(310)[001] Cu grain boundary by direct thermodynamic integration, or measure the 300 K site occupancy by atom-probe tomography: if the order is not vertices first, then near-vertex sites, then bottom sites, the chemical-potential functional is not capturing equilibrium segregation.","supporting_citations":[],"review_version":1}