{"id":"d662e936-726a-4c50-a7b9-adc48bf0c401","arxiv_id":"2501.11101","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Nickel atoms prefer to segregate into interstitial cavities within kite-like grain boundary structures in aluminum, and a new Voronoi-based method with machine learning can identify and predict these sites.","lead":"Simulations of an aluminum-nickel alloy show that nickel atoms can slip into empty cavities inside grain boundaries, not just replace aluminum atoms as usually assumed. This could make predictions of grain boundary segregation and alloy stability more accurate, and gives designers a computational tool to find such sites.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim rests on the Purja Pun–Mishin EAM potential's relative energetics for Ni at interstitial versus substitutional Al GB sites; alternate-potential checks cover only one bicrystal, so potential fidelity remains the key open risk.","rationale":"The reader's weakest_assumption identifies exactly the concern I consider most load-bearing: the fidelity of the primary EAM potential for interstitial versus substitutional Ni energetics. All physical and ML conclusions inherit this potential's energy ranking, so if that ranking is wrong, the central claim collapses regardless of the paper's strong internal consistency. The manuscript's own mitigation—testing two additional EAM potentials on a single Σ5(210) bicrystal—is helpful but not decisive, because those are also empirical EAM potentials and may share systematic biases for interstitial energetics. The other concerns raised by the reader (threshold circularity, within-model ML, partial data release) are secondary: they affect the practical tools and quantitative predictions, but the potential-dependence question strikes at the existence of the phenomenon itself. A focused DFT benchmark at two representative GBs would settle the existential question with modest effort. Since the paper already merits conditional acceptance pending such validation, my assessment does not change the reader's verdict.","tokens_in":23295,"tokens_out":1850,"duration_ms":22630,"concrete_test":"Perform DFT (VASP or equivalent, PAW-PBE, spin-polarized) calculations of the single-Ni substitutional and interstitial segregation energies, defined as in Eq. (1), at the kite-core and kite-tip candidate sites of the Σ5(210) and Σ5(310) Al STGBs, using the same relaxed GB supercells as in Section 3.1. Compare the DFT relative ordering (interstitial minus substitutional segregation energy) with the Purja Pun–Mishin EAM predictions at the same sites. If DFT gives E_substitutional ≤ E_interstitial at sites where EAM predicts strongly negative interstitial segregation, the central phenomenon is potential-dependent and the broad claims are not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central observation—that Ni preferentially occupies interstitial kite-core sites in Al GBs—is generated entirely by hybrid MD/MC simulations using the Purja Pun–Mishin EAM potential (Ref. [77]). Every downstream result depends on this potential: the 174-bicrystal survey, the per-site segregation energies in Eqs. (1) and (3), the NC segregation spectra, and the SOAP-ML training labels. The physical claim is that interstitial segregation is a real, general phenomenon in Al–Ni, not an artifact of one empirical potential. The manuscript partially addresses this in Section 6.3 and Supplementary Fig. S15 by repeating the Σ5(210) hybrid simulation with two additional EAM potentials [93,94], which do reproduce interstitial segregation. However, those potentials are also empirical EAM fits to similar datasets, so they are not independent benchmarks for the interstitial-vs-substitutional energy ordering. The broader claims across many GB types and the quantitative segregation energies rest on a single potential's ranking. Prior first-principles work [68–71] supports interstitial Cu in Al GBs, but Ni is not the same solute and those studies are zero-temperature, so they do not directly validate the EAM Ni energetics. The risk is not internal inconsistency but external correctness: if the EAM potential over-stabilizes interstitial Ni relative to substitutional Ni at kite sites, the phenomenon itself could be a simulation artifact, and the ML predictor would simply learn that artifact.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript reports hybrid molecular dynamics/Monte Carlo simulations of Al-Ni bicrystals and nanocrystalline samples, showing that Ni atoms preferentially occupy interstitial sites (rather than substitutional sites) in the kite-like cores of numerous CSL grain boundaries at 300 K. The authors classify the behavior into intraplanar and interplanar interstitial segregation, observe segregation-induced GB transitions (kite transitions and nano-faceting), and develop a Voronoi/DBSCAN-based method to identify interstitial candidate sites. They use this method to compute per-site interstitial segregation energies in a nanocrystalline Al sample, incorporate them into a dual-solute segregation model, and train linear regression models with SOAP descriptors to predict these energies. The central claim is that interstitial segregation is a general phenomenon in substitutional alloys, challenging the common assumption that substitutional solutes occupy only substitutional sites.","tokens_in":23544,"tokens_out":6572,"duration_ms":64547,"significance":"If correct, the finding would require grain boundary segregation models to include an interstitial occupancy channel, with consequences for predictions of GB chemistry, stability, and transitions. The paper provides a practical site-identification tool with publicly available code, applies it to both bicrystals and nanocrystalline samples, and demonstrates that SOAP-based linear regression can predict interstitial segregation energies with R-squared values around 0.83-0.85. The central observation is supported by atom-density comparisons, repeated tests at 400-500 K, and a repeat of the Sigma5(210) simulation with two additional EAM potentials. However, the physical generality is currently tied to a single interatomic potential, and the site-identification method's validation includes a tuning-to-simulation component, which tempers the strength of the broader claims.","major_comments":[{"comment":"The broad claims of generality in the abstract and title rest on the Purja Pun-Mishin EAM potential (Ref. [77]) for the Al-Ni system. In Section 6.3 and Supplementary Fig. S15, the additional EAM potentials are tested only for the single Sigma5(210) bicrystal, and all other results--including the 174-GB survey, the segregation energies of Eqs. (1) and (3), and the ML training labels--are generated with that one potential. Because the central physical claim is that interstitial segregation is a real, general phenomenon, the authors should provide a concrete benchmark of the interstitial-vs-substitutional energy ranking against first-principles calculations for Ni in the kite cores of at least the three representative GBs (Sigma5(210), Sigma9(221), Sigma11(332)) and one ATGB, or run the hybrid MD/MC test with independent potentials on a diverse set of GBs and demonstrate that the occupancy ranking is unchanged.","section":"Section 6.3 / Supplementary Fig. S15"},{"comment":"The validation of the interstitial site identification method in Section 4.2 is partly circular. The DBSCAN parameters d_c=1.25 A and N_min=5-7, the free-volume thresholds (10.8 A^3 for bicrystals and 10.0 A^3 for the NC sample), and the distance filter d_min=2.0 A are selected in Section 4.1 so that the identified sites match the hybrid MD/MC segregation patterns in Figs. 2 and 4. The subsequent comparison in Fig. 6 therefore compares the method's output with the very simulations used to choose these thresholds. To support the claim that the method is robust, the parameters should be fixed on a subset of GBs and then evaluated on held-out GBs, or the identified sites should be cross-checked against an independent method such as DFT-relaxed interstitial occupancy.","section":"Section 4.1 / Fig. 5"},{"comment":"The claim in Section 5.2 that the full segregation-energy dataset 'significantly enhances' the accuracy of GB segregation predictions is not quantified. Figure 8(b) shows only visual agreement between the DS, DS-SS-int, and full curves and the hybrid MD/MC points adapted from Ref. [66]; no numerical error metric (e.g., RMSD or mean absolute deviation in predicted vs simulated GB solute concentration) is given. Please provide a quantitative comparison and state whether the improvement is statistically meaningful. Because the MD/MC reference data are also produced with the same EAM potential, the comparison is a test of the model within that potential, not an experimental validation.","section":"Section 5.2 / Fig. 8(b)"}],"minor_comments":[{"comment":"There is a typo in 'lang-range elastic interactions'; it should be 'long-range'. The explanation of E_ref^Al,bulk as balancing the atom count is terse; please clarify that inserting a Ni atom adds a particle and the reference energy accounts for removing one Al atom.","section":"Section 2.2 / Eq. (1)"},{"comment":"The definition of interstitial segregation as occurring 'without significantly altering their original structures' is difficult to reconcile with the later description of interstitial-segregation-induced kite transitions and faceting in Section 6.2. Please clarify whether the transition cases are meant to be excluded from the definition or treated as a separate category.","section":"Section 3.1 / Fig. 2"},{"comment":"The ML predictions use SOAP descriptors, but it is not stated explicitly whether the descriptors are computed on the unrelaxed identified site geometry or on the relaxed geometry after the molecular statics relaxation. Given the average displacement of 0.139 A discussed in the same section, this distinction matters for reproducibility and should be specified.","section":"Section 6.1 / Fig. 9"},{"comment":"The sentence 'The parameters can be adjusted as needed' is vague; the paper does list tested ranges for d_c and N_min, but a brief sensitivity analysis (e.g., number of identified sites vs d_c) would make the method more reproducible and help readers understand the robustness of the 1.25 A and N_min=5-7 choices.","section":"Section 4.1 / Fig. 5"},{"comment":"The comparison for the Sigma5(9 13 0)/(310) ATGB uses hybrid MD/MC simulations at a solute concentration of 0.5 at.%, while the site identification is performed on the pure GB. If the GB structure evolves during segregation, please confirm that the comparison between the identified sites and the segregated-Ni distribution is still meaningful and state any limitations.","section":"Section 4.2 / Fig. 6"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is worth reading if you work on grain boundary segregation or nanocrystalline alloy stability. The genuinely new content is the systematic survey of Ni interstitial segregation in Al grain boundaries across 174 CSL bicrystals, the enhanced Voronoi/DBSCAN method for identifying interstitial candidate sites for metallic solutes (as opposed to hydrogen), and a linear ML model, trained with SOAP descriptors, that predicts per-site interstitial segregation energies.\n\nThe paper does several things well. The hybrid MD/MC simulations are internally consistent: Ni atoms occupy kite-core hollow sites, atom density at the GB increases, and the qualitative result holds at 400–500 K and with two additional EAM potentials in the Σ5(210) case. The downstream test—identifying sites in a nanocrystalline sample, computing segregation energies, and using them in a spectral model to predict GB concentration as a function of total concentration—is a meaningful validation that goes beyond the bicrystal illustrations.\n\nThe soft spots are real but not fatal. The site identification thresholds are chosen to reproduce the same hybrid MD/MC segregation patterns used for validation, so the method's accuracy is partly circular; the ML models train and test on energies from the same interatomic potential, so they show within-model transferability across structures, not physical transferability; the claimed improvement in segregation prediction is shown visually without a numerical error metric; and only the site identification code is released, not the full datasets.\n\nThe load-bearing assumption is the Purja Pun–Mishin EAM potential's relative energetics for Ni at interstitial versus substitutional sites. The stress-test note is fair: the alternate-potential checks cover only one bicrystal and are also empirical EAM fits, so they are not independent benchmarks. That said, the physical rationale (undersized solute in open kite cores) is plausible, and prior first-principles studies plus experimental observation for Cu in Al support the general phenomenon. I would not sink the paper on this, but it should be flagged prominently.\n\nThe paper deserves a serious referee. For revision, the authors should add quantitative error metrics for the prediction improvement, release the segregation energy datasets, and ideally benchmark a few representative GB sites with DFT to test the EAM potential's energy ranking.\n\nRecommendation: engage with it. The tool and the dataset will be useful to the community regardless of whether the potential's energy ranking survives DFT scrutiny.","headline":"A systematic, useful computational study of Ni interstitial segregation in Al grain boundaries; the site-identification tool is a real contribution, but the EAM potential fidelity warrants scrutiny in review.","tokens_in":24171,"tokens_out":3806,"would_cite":true,"duration_ms":36638,"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":"Nickel atoms segregate into open pockets inside aluminum grain boundaries, not just onto lattice sites.","keywords":["grain boundary segregation","interstitial segregation","substitutional binary alloys","Al-Ni system","hybrid MD/MC simulation","Voronoi site identification","nanocrystalline alloys","SOAP descriptors"],"falsifier":"A density-functional-theory calculation of a single Ni atom at the kite-core interstitial site versus the kite substitutional site in the Σ5(210) boundary would settle the matter; if the interstitial site is not thermodynamically preferred at 300 K, the observed pattern would likely be a classical-potential artifact.","tokens_in":22995,"feed_emoji":"⚛️","tokens_out":5385,"duration_ms":50842,"temperature":0.7,"pith_summary":"This paper argues that solute atoms in a substitutional alloy can segregate to grain-boundary interstitial sites, not only to substitutional lattice sites, and demonstrates the phenomenon in the Al-Ni system through hybrid molecular dynamics/Monte Carlo simulations at 300 K. Ni atoms, slightly smaller than Al, preferentially fill hollow spaces inside kite-like grain-boundary structural units, and this interstitial occupancy can change the boundary structure itself, producing kite transitions and nano-faceting. The authors build a practical method for finding such sites—Voronoi-cell vertices merged by clustering and filtered by free volume and atomic-distance criteria—and show that including interstitial segregation energies improves predictions of grain-boundary solute content in nanocrystalline Al. If this holds, grain-boundary segregation models for substitutional alloys must add an interstitial channel alongside the usual substitutional one.","feed_headline":"Ni atoms slip into open pockets inside Al grain boundaries","feed_subtitle":"Simulations find interstitial segregation in a substitutional alloy, improving grain-boundary segregation predictions.","key_machinery":"The load-bearing tool is an enhanced Voronoi-based interstitial-site identification method. Voronoi polyhedra are built around every atom, and their vertices—points of maximal hollow space—are merged into candidate sites with a DBSCAN clustering algorithm; candidates are then filtered by free volume (above roughly the atomic volume of Ni) and by minimum distance to GB atoms, so that only sites large enough to host a metallic solute survive. This site list feeds molecular-statics calculations of single-solute and dual-solute interstitial segregation energies, and the energies are combined with a spectral dual-solute model to predict grain-boundary concentrations. A linear regression on SOAP (smooth overlap of atomic positions) descriptors of the local environment then predicts per-site interstitial segregation energies, transferring from a 16-nm nanocrystalline sample to a larger one.","core_discovery":"The central claim is that grain-boundary interstitial segregation is a real and common mode of solute accommodation in substitutional binary alloys, contrary to the usual assumption that segregated solutes replace solvent atoms at lattice sites. Using hybrid MD/MC simulations, the paper finds that Ni atoms in Al preferentially occupy the open cores of kite-like structural units in a wide range of Σ3, Σ5, Σ9, Σ11, and Σ13 boundaries and in nanocrystalline grain-boundary networks, forming intraplanar and occasionally interplanar segregation patterns. The authors further show that these interstitial solutes can drive structural transitions at room temperature, including kite flipping and nano-faceting, and that accounting for interstitial segregation energies materially improves the match between spectral segregation models and simulated grain-boundary concentrations.","pith_inferences":["The same mechanism should operate in other FCC-based substitutional alloys where the solute is noticeably smaller than the solvent, such as Cu in Al, Co in Al, or Fe in Al; the paper's own Ag-Cu and Ta-Cu tests point this way.","If interstitial occupancy is as widespread as suggested, experimental atom-probe or STEM studies of decorated Al grain boundaries should look for solute atoms that sit off the substitutional lattice, a signature that is currently often interpreted as a measurement artifact.","The free-volume threshold (~10.8 ų in bicrystals, 10.0 ų in nanocrystalline samples) is likely transferable across FCC solutes with similar atomic radii, but would need re-calibration for larger or smaller solutes.","A direct density-functional-theory benchmark of interstitial versus substitutional segregation energies at a few kite cores would decide how much of the phenomenon is a real energetic preference and how much is an artifact of classical potentials."],"forward_implications":["Grain-boundary segregation models for substitutional alloys should include interstitial occupancy as a distinct channel, not only substitutional site swapping.","Kite-like structural units are a necessary-but-not-sufficient condition for interstitial segregation; loose-packed boundaries such as Σ13(510) still favor substitutional sites.","Room-temperature segregation can restructure grain boundaries, producing kite transitions and nano-faceting that would be missed by zero-temperature first-principles site ranking.","Adding interstitial segregation energies to spectral models improves predictions of grain-boundary solute concentration in nanocrystalline Al-Ni, with dual-solute interstitial energies giving the largest gain.","SOAP-based linear models trained on small nanocrystalline samples can predict per-site interstitial segregation energies in larger samples, enabling screening without full molecular-statics calculations."],"supporting_citations":[{"why":"Supplies the Al-Ni EAM potential used for all hybrid MD/MC, molecular-statics energies, and ML labels.","marker":"[77]"},{"why":"Provides the variance-constrained semi-grand-canonical hybrid MD/MC algorithm used to simulate segregation at 300 K.","marker":"[78]"},{"why":"Provides the spectral segregation-energy learning framework and the nanocrystalline Al dataset used for comparison and ML transfer.","marker":"[55]"},{"why":"The authors' earlier dual-solute model provides the substitutional segregation-energy spectra and the reference DS model that interstitial energies are added to.","marker":"[66]"},{"why":"The Voronoi-based interstitial-site method for hydrogen in nanocrystalline Pd that this paper extends to larger metallic solutes.","marker":"[85]"},{"why":"Supplies the 174 CSL grain-boundary bicrystal models used for the systematic survey.","marker":"[73]"},{"why":"First-principles evidence of Cu preferring hollow interstitial sites in Al boundaries, the prior observation this work generalizes.","marker":"[71]"},{"why":"Introduces SOAP descriptors used to represent local atomic environments for ML prediction of interstitial segregation energies.","marker":"[82]"},{"why":"Provides the DBSCAN clustering and linear-regression implementations used for site merging and ML training.","marker":"[84]"}],"fun_headline_variants":["Ni atoms wedge into Al boundary gaps","Simulations show Ni fills GB interstices in Al","Unexpected Ni interstitial sites in Al boundaries","Ni takes open spots, not lattice, in Al GBs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire picture rests on one classical interatomic potential for Al-Ni: if it mis-ranks the energy of a Ni atom in an interstitial site versus a substitutional site at an aluminum grain boundary, the phenomenon could be a simulation artifact rather than real physics.","fun_headline_variants_meta":{"raw":{"variants":["Ni atoms wedge into Al boundary gaps","Simulations show Ni fills GB interstices in Al","Unexpected Ni interstitial sites in Al boundaries","Ni takes open spots, not lattice, in Al GBs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000141,"raw_usage":{"total_tokens":1143,"prompt_tokens":904,"completion_tokens":239,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":178}},"tokens_in":520,"tokens_out":239,"duration_ms":3631,"temperature":1.0,"reasoning_tokens":178,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:38:36.461485+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A density-functional-theory calculation of a single Ni atom at the kite-core interstitial site versus the kite substitutional site in the Σ5(210) boundary would settle the matter; if the interstitial site is not thermodynamically preferred at 300 K, the observed pattern would likely be a classical-potential artifact.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"First-principles evidence of Cu preferring hollow interstitial sites in Al boundaries, the prior observation this work generalizes."}],"review_version":1}