REVIEW 4 major objections 4 minor 4 references
Analytic model for grain-boundary segregation ener-gies in metal polycrystal
T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash
Pith's one-line read The paper claims that grain-boundary segregation energies in metal polycrystals are determined by a strained coordination number, with a closed-form formula matching simulation data to about 5 kJ/mol.
desk verdict Useful empirical descriptor, but the 'prediction' claim rests on fitting the same MD data, so treat Eq. (7) as an interpolation formula until out-of-sample tests appear. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the strained coordination number CN_str = sum_i (d_0 / d_i), a dimensionless measure of how many first neighbors an atom has and how stretched their bonds are, with d_0 the bulk bond length and d_i the actual bond length to neighbor i. Stretched bonds count less than intact bonds, so CN_str unifies plastic strain and bond breaking into one energy descriptor. The argument shows that segregation energy is linear in CN_str within coordination-number intervals, then separates variables: the slope scales linearly with CN, with solute radius r_s, and with 1/r_m, while the matrix cohesive energy E_c,m sets the overall scale. Collecting these proportionalities yields Eq. (7), with a neighbor-counting cutoff of 1.15 d_0 chosen for optimal linearity.
What would settle it
Run first-principles or careful molecular-dynamics segregation-energy calculations for a solute-matrix pair outside the 16-element set and check whether energies over a wide range of boundary sites fall on the straight line E_seg = k (CN_str - CN) - 10 with k = (15/2) E_c,m (30 - CN)(r_s/r_m - 1) + 10; a systematic deviation beyond the reported few kJ/mol would break the model.
Extended reading notes
Core claim
The central claim is that the segregation energy E_seg of a solute at a polycrystalline grain boundary is given by an analytic formula involving only the strained coordination number CN_str, the usual coordination number CN, the solute and matrix atomic radii r_s and r_m, and the matrix cohesive energy E_c,m: E_seg = [(15/2) E_c,m (30 - CN) (r_s/r_m - 1) + 10] (CN_str - CN) - 10. The authors report mean absolute errors of about 5 kJ/mol over a segregation-energy span of about 420 kJ/mol, with per-pair errors generally below 6.5 kJ/mol and often near 1 kJ/mol. This accuracy is comparable to machine-learning models, but the analytic form exposes the physical coupling: oversized solutes prefer stretched, low-CN_str sites, undersized solutes prefer compressed, high-CN_str sites, and only first-nearest-neighbor bonding matters.
Load-bearing premise
The model assumes that a boundary atom's energy is the simple sum of pairwise first-neighbor bond energies, each falling off as one over the stretched bond length, so that the strained coordination number is the right energy descriptor.
Editorial extensions
If this is right
- For the studied metals, segregation energies can be computed analytically from structural and elemental data, with mean absolute errors around 5 kJ/mol over a 420 kJ/mol range.
- Oversized solutes are predicted to prefer stretched, low-CN_str boundary sites; undersized solutes prefer compressed, high-CN_str sites.
- The descriptor transfers to bulk metallic glasses, supporting the view that polycrystal grain boundaries are amorphous-like in their bonding.
- Only first-nearest-neighbor effects matter, so the solute–matrix coupling at polycrystal grain boundaries is strongly screened and localized.
- The formula supplies a fast screening tool for alloy design, since each site only requires a neighbor list rather than a full simulation or a trained model.
Reading between the lines
- A testable extension is to apply Eq. (7) to alloy systems outside the 16 elements by plugging in only radii and cohesive energy, and checking whether the reported few-kJ/mol accuracy persists without any refitting.
- Because Eq. (7) is a zero-temperature configurational energy, finite-temperature segregation predictions would still need entropy and vibrational corrections; the analytic energy is only one input to an isotherm.
- The 30 - CN factor and the 15/2 prefactor likely reflect a geometric packing relation; deriving them from first principles could extend the model to non-metallic or multicomponent boundaries.
- The locality claim could be probed by comparing CN_str predictions against explicit second-neighbor descriptors on the same polycrystal data to see where the first-neighbor approximation starts to fail.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes an analytic model for solute segregation energies at grain boundaries (GBs) in metal polycrystals. The central quantity is a 'strained coordination number' CNstr = sum(d0/d_i) within a first-neighbor cutoff of 1.15 d0, motivated by a Coulombic-like 1/d bond-energy assumption. The authors report that segregation energy Eseg is linear in CNstr for each solute-matrix pair, and that the slope and intercept of these lines depend linearly on CN, solute radius r_s, matrix radius r_m, and matrix cohesive energy E_c,m. These correlations are assembled into Eq. (7), Eseg = [(15/2) E_c,m (30-CN)(r_s/r_m - 1) + 10](CNstr - CN) - 10. The paper claims mean absolute errors below about 5 kJ/mol against >2x10^7 molecular-dynamics data points from Ref. 14, comparable to machine-learning models, and shows that segregation energies in bulk metallic glasses (BMGs) scale linearly with CNstr using DFT data. The authors conclude that bonding at polycrystal GBs and BMGs is Coulombic-like and localized, and that Eq. (7) is an effective tool for alloy design.
Significance. If the model were validated as a genuine predictor for unseen solute-matrix pairs, it would be a practically valuable and physically transparent alternative to black-box machine-learning models: it requires only CN, CNstr, r_s/r_m, and E_c,m, and its analytic form can be evaluated instantly for large polycrystal structures. The paper's strengths are its explicit closed-form expression, the documentation of linear relations with R^2 values, and the use of external DFT data for BMGs, which provides some out-of-sample evidence for the linearity of Eseg versus CNstr. However, the central predictive claim is not yet established because the constants in Eq. (7) are fit to the same Ref. 14 dataset used for the reported accuracy, and the model has an unphysical limiting behavior for a solute identical to the host. The paper is therefore a promising framework that needs additional validation and correction before the claimed generality can be accepted.
major comments (4)
- [Fig. 3d-f and Methods ('Polycrystal structures')] The reported MAEs of 5.07 kJ/mol overall and below 6.5 kJ/mol per pair are training errors, not prediction errors. The linear relations in Eqs. (2)-(6) and the constants in Eq. (7) are regressed on the MD segregation data of Ref. 14, and Fig. 3d-f compares Eq. (7) against that same dataset. No holdout set, cross-validation, or independent polycrystal dataset is used. The central claim of predictive capability across solutes and matrices therefore requires a genuinely out-of-sample test, for example by fitting on a subset of solute-matrix pairs and validating on held-out pairs or on independently computed polycrystal data.
- [Eq. (7) and paragraph following it] The model has an unphysical limiting behavior when the solute is chemically identical to the host. Setting r_s = r_m in Eq. (7) gives Eseg = 10(CNstr - CN) - 10, so at a site with CNstr = CN the predicted segregation energy is -10 kJ/mol, whereas the segregation energy of a host atom replacing itself must be zero by definition. More generally, Eq. (7) predicts Eseg = -10 for every solute at any unstrained GB site with CNstr = CN, which contradicts the meaning of Eseg as the energy difference between solute at a GB site and in the bulk. This issue points to the fitted intercept in Eq. (4) being an artifact of the dataset rather than a physically required constant, and it must be corrected before the model can be regarded as general.
- [Eq. (7) and its derivation from Eqs. (3)-(6)] The numerical constants 15/2, 30, +10, and -10 in Eq. (7) are fitted parameters, but the paper presents them without derivation, uncertainty estimates, or a sensitivity analysis. The cutoff radius of 1.15 d0 is likewise selected for optimal linearity ('optimal accuracy at 1.15d0'), and the supplementary material only states that the linear relation is 'weakly affected' by the cutoff. Without error bars on these constants or a demonstration that the predictions are stable with respect to the cutoff and to the fitting procedure, the claimed universality of Eq. (7) is not quantified. The authors should report the fitted values with uncertainties and test the sensitivity of the MAEs to the cutoff and to the regression ranges.
- [Eq. (7) and Fig. 2] Solute chemistry enters Eq. (7) only through the metallic radius r_s, and matrix chemistry only through r_m and E_c,m. Consequently, the model predicts identical segregation energies for all solutes with the same metallic radius in a given matrix, regardless of their electronic structure or chemical identity. This degeneracy is not discussed, and the paper provides no evidence that it is physical. Known chemical effects, such as d-band filling or electronegativity differences, are absent, so the authors should either justify the degeneracy with data for same-radius solutes in a common matrix or acknowledge this as a limitation of the model's transferability.
minor comments (4)
- [Main text, paragraph after Eq. (7) and elsewhere] There is a typo 'BGMs' where 'BMGs' is intended, and in the Methods the phrase 'ab intio MD' should read 'ab initio MD'.
- [Methods equations] The Methods section numbers its segregation-energy formulas as Eqs. (1) and (2), but Eq. (1) in the main text is the definition of CNstr. The duplicate equation numbering should be corrected for clarity.
- [Data availability] The text states that the authors 'compile CNstr into a Python code,' but the Data Availability section only offers data upon request and does not mention code availability. Providing the code would substantially help readers apply Eq. (7) to new polycrystal structures.
- [Figure 1 caption] The caption ends with 'Pd solute in Mg matrix. 14', which appears to contain a stray citation or reference marker. This should be cleaned up.
Circularity Check
No genuine out-of-sample test: Eq. (7)'s constants are fit to the same Ref. 14 MD data used for the reported MAEs, so the cross-solute/matrix prediction claim is not yet established.
-
fitted input called prediction
[The predictive-accuracy paragraph after Eq. (7), shown in Figs. 3d-3f.]
"We now study the predictive accuracy of our scheme in determining the segregation energies of solutes at polycrystal GBs and BMGs, by comparing with >2×107 MD data points and >1000 DFT data points. If one compares all the studied solute-matrix pairs together, the MAEs of our scheme is 5.07 kJ/mol with the segregation-energy spans about 420 kJ/mol (Fig. 3d)."
The MD data are entirely from Ref. 14, and the constants in Eq. (7) are obtained by the regression chain in Eqs. (3)-(6) on those same Eseg-versus-CNstr data via per-pair and per-CN-interval fits of k and b. No held-out split or independent test set is described. Therefore Fig. 3d-f and the quoted 5.07 kJ/mol MAE compare Eq. (7) with its own fitting data; by construction these are in-sample training errors, not predictions across solutes and matrices.
-
fitted input called prediction
[Equation (7) and Fig. 3a of the main text.]
"The predicted slope values by the term 15/2 Ec,m(30-CN)(rs/rm -1)+10 of Eq. (7) are in good agreement with the calculated ones (see Fig. 3a), demonstrating the robustness of our model."
The calculated slopes are the ks-m values from fitting Eq. (2) to the Ref. 14 MD segregation energies, while the slope expression inside Eq. (7) was itself obtained by regressing those same ks-m against CN, rs/rm, and Ec,m (Eqs. (3)-(6)). Fig. 3a thus correlates the formula with the fit used to construct it; it is a self-consistency check, not an independent validation of the slope formula.
full rationale
The central claim is an analytic predictor, but Eq. (7) is assembled from empirical regressions over the same Ref. 14 MD data used for the reported accuracy figures, so the MAEs in Figs. 3d-3f are in-sample. Fig. 3a is likewise a fit-versus-fit correlation. The BMG DFT results (Figs. 3b-3c) are genuinely external and support the CNstr linearity descriptor for Pt-Fe and Ag-Mg, but they do not test the absolute transferability of Eq. (7)'s constants across the 16-element solute/matrix space. No load-bearing self-citation was found: Ref. 26 is cited only as a contrast for symmetric tilt GBs, and no uniqueness theorem is imported. Overall this is partial circularity in the core prediction claim, not full equivalence.
Assumptions & free parameters
free parameters (4)
- Cutoff radius factor for first-nearest-neighbor counting =
1.15 d0
- Slope prefactor in Eq. (7) =
15/2
- Coordination offset in Eq. (7) =
30
- Intercept and slope offset constants in Eq. (7) =
+10 and -10 (units unspecified)
assumptions (3)
- domain assumption Bond energy at polycrystal GBs is pairwise additive and proportional to 1/d_i for first nearest neighbors only.
- domain assumption The MD segregation-energy dataset of Ref. 14 (LAMMPS EAM potentials at 0 K) is an accurate ground truth for segregation energies in polycrystals.
- ad hoc to paper The empirical linear relations (Eqs. 2-6) and the fitted constants remain valid outside the fitted intervals and for unstudied solute-matrix pairs.
Cite this review
Pith. "Pith review of Analytic model for grain-boundary segregation ener-gies in metal polycrystal." pith.science (2026). https://pith.science/paper/F2S2ZGAR
@misc{pith2026241216466,
author = {Pith},
title = {Pith review of: Analytic model for grain-boundary segregation ener-gies in metal polycrystal},
year = {2026},
howpublished = {\url{https://pith.science/paper/F2S2ZGAR}},
note = {Machine review of arXiv:2412.16466}
}
read the original abstract
Solute segregation at grain boundaries (GBs) of polycrystals strongly impacts the mechanical properties of metals including strength, fracture, embrittlement, and corrosion. However, the complexity of GB structures and the large chemical space of solutes and matrices impede the understanding of segregation. Herein, we identify a physical-based determinant, by unifying the effects of plastic strain and bonding breaking, for determining the segregation energies at GBs. By further combining with the usual coordination number, atomic radius of solutes and matrices, and cohesive energy of matrices, we build an analytic framework to predict segregation energies of polycrystal GBs across various solutes and matrices. These findings indicate an unusual Coulombic-like and localized nature of the bonding at polycrystal GBs and bulk metallic glasses (BMGs). Our scheme not only uncovers the coupling rule of solutes and matrices for GB segregation in polycrystals, but also provides an effective tool for the design of high-performance alloys.
Figures
Reference graph
Works this paper leans on
-
[27]
Thompson, A. P. et al. LAMMPS-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Comput. Phys. Commun. 271, 108171 (2022)
work page 2022
-
[28]
Atomsk: A tool for manipulating and converting atomic data files
Hirel, P. Atomsk: A tool for manipulating and converting atomic data files. Comput. Phys. Commun. 197, 212-219 (2015)
work page 2015
-
[29]
Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169-17979 (1996)
work page 1996
-
[30]
Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple,. Phys. Rev. Lett. 77, 3865-3868 (1996). Competing interests The authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/xxx. Acknowledgments The authors are thankful for the support from t...
work page 1996
Reviewed August 11, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.