pith. sign in

arxiv: 2606.19798 · v1 · pith:73RVXHPHnew · submitted 2026-06-18 · ❄️ cond-mat.mtrl-sci

MinSurf: resolving the atomic-scale stability landscape of mineral surfaces

Fengzijun Pan , Zhoulin Liu , Pingyang Zhang , Jiaqiu Xu , Zepeng Fan , Dawei Wang , Jianzhong Pei This is my paper

Pith reviewed 2026-06-26 16:43 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci
keywords mineral surfacessurface energymachine learning potentialWulff constructionhigh-throughput computationdensity functional theorycrystal morphologyatomic termination
0
0 comments X

The pith

MinSurf trains a neural potential on 764 DFT mineral slabs to predict surface energies at 0.0119 eV/Ų error and 11400x speedup.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents MinSurf as a workflow that enumerates atomic terminations of mineral surfaces, computes their energies with DFT on a training set of 764 slabs from ten minerals, fits a machine-learned potential called MinNEP, and applies Wulff construction to obtain equilibrium shapes. This is motivated by the fact that full DFT across all possible terminations is too costly, so most simulations pick facets without systematic comparison. A reader would care because surface terminations control reactivity in carbon mineralization, geo-energy, catalysis, and contaminant processes, and having representative low-energy models changes what simulations can be trusted. The work reports that MinNEP reproduces the DFT energy ordering and the dominant exposed facets while passing an X-ray diffraction check on alpha-quartz.

Core claim

MinSurf combines surface enumeration, DFT labeling of 764 slabs plus 90 bulk references, training of the MinNEP interatomic potential, and Wulff construction to resolve which atomic terminations are stable, to map surface-energy landscapes, and to predict equilibrium morphologies; the resulting potential matches DFT surface energies to a mean absolute error of 0.0119 eV per square angstrom, delivers a 1.14 times 10 to the 4 acceleration, and preserves the morphology-determining energy hierarchy.

What carries the argument

MinNEP, a machine-learning interatomic potential trained on DFT surface energies of enumerated slabs from ten minerals, used to evaluate energies across many more terminations than direct DFT allows.

If this is right

  • Surface models for simulations of carbon mineralization and heterogeneous catalysis can be chosen by predicted energy rather than convention.
  • The DFT-derived ordering of surface stabilities is preserved at the accelerated scale.
  • Dominant facets in equilibrium morphologies match those from direct DFT.
  • X-ray diffraction patterns provide an independent check that the chosen reference structures are crystallographically consistent.
  • Reproducible atomic-scale surface models become available for high-throughput studies of mineral interfaces.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same enumeration-plus-potential approach could be applied to additional mineral families or to surfaces with defects if the potential remains accurate outside the training distribution.
  • Morphology predictions could be tested against crystal-growth experiments that control for temperature and supersaturation rather than only against room-temperature XRD.
  • The framework supplies a concrete route to compare many more candidate terminations than current DFT budgets allow, which may shift practice from facet-level to termination-level modeling in interface studies.

Load-bearing premise

The neural potential trained on the 764 labeled slabs will give accurate energies for atomic terminations it never saw during training.

What would settle it

Experimental measurement of surface terminations or crystal facet distributions for one of the ten minerals that differs from the MinNEP-predicted Wulff shape.

read the original abstract

Mineral surfaces govern interfacial reactivity in carbon mineralization, geo-energy storage, contaminant immobilization, heterogeneous catalysis and electrochemical interface engineering. Yet atomistic simulations often rely on commonly used facets or facet-level stability criteria, while distinct atomic terminations of the same crystallographic orientation are rarely resolved systematically because experimental characterization and density functional theory (DFT) calculations remain costly across large surface spaces. Here we present MinSurf, a high-throughput framework that resolves mineral surface selection as a surface-energy and morphology problem. MinSurf integrates surface enumeration, DFT labelling, machine-learning interatomic potentials and Wulff construction to predict stable terminations, surface-energy landscapes and equilibrium crystal morphologies. Applied to ten representative minerals, MinSurfSet comprises 764 surface slabs, with 90 corresponding oriented unit cells constructed as bulk references for surface-energy evaluation. The resulting MinNEP model predicts DFT surface energies with a mean absolute error of 0.0119 eV per Angstrom squared and achieves an overall acceleration of 1.14 x 10^4 relative to DFT. MinNEP preserves the DFT-derived morphology-determining surface-energy hierarchy and reproduces the dominant Wulff-exposed facets, while X-ray diffraction provides an independent crystallographic consistency check for alpha-quartz benchmark. By linking atomic terminations, surface energies and equilibrium morphologies, MinSurf provides reproducible and physically representative surface models for high-throughput simulations of mineral interfaces across energy, environmental and advanced inorganic materials.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents the MinSurf framework, which enumerates atomic terminations for mineral surfaces, labels 764 slabs from ten minerals with DFT, trains the MinNEP machine-learning interatomic potential, and applies Wulff construction to predict stable surface energies, hierarchies, and equilibrium morphologies. It reports that MinNEP achieves a mean absolute error of 0.0119 eV/Ų relative to DFT, a 1.14×10^4 speedup, preserves the DFT-derived surface-energy ordering, reproduces dominant Wulff facets, and is consistent with X-ray diffraction for the α-quartz benchmark.

Significance. If the reported performance and hierarchy preservation hold under proper validation, MinSurf would provide a scalable, reproducible route to mapping atomic-scale surface stability landscapes for minerals relevant to carbon mineralization, geo-energy, and catalysis. The integration of enumeration, DFT labeling, ML surrogate, and Wulff construction, together with the concrete acceleration factor, represents a practical advance for high-throughput interfacial modeling.

major comments (3)
  1. [Abstract] Abstract (MinNEP performance paragraph): The MAE of 0.0119 eV/Ų is stated without any description of the train/validation/test partitioning, held-out terminations, or leave-one-mineral-out protocol. Because the central claim is that MinNEP reproduces the DFT surface-energy hierarchy across the full enumerated space, the absence of explicit generalization metrics on unseen terminations is load-bearing.
  2. [Abstract] Abstract and results on morphology: The claim that MinNEP 'preserves the DFT-derived morphology-determining surface-energy hierarchy' and 'reproduces the dominant Wulff-exposed facets' is presented without uncertainty quantification or error bars on the predicted energies used for ranking. This leaves open whether small extrapolation errors could alter facet ordering.
  3. [Methods] Methods (MinNEP training description): No details are supplied on how the 764 DFT slabs were split for training versus testing, nor on whether any enumerated terminations lie outside the training distribution (different coordination or stoichiometry). This directly affects the reliability of applying the surrogate to the complete surface space.
minor comments (2)
  1. [Abstract] The abstract mentions '90 corresponding oriented unit cells' but does not clarify how these bulk references were chosen or whether they introduce any systematic bias in surface-energy evaluation.
  2. [Figures] Figure captions and text should explicitly state whether the reported MAE and hierarchy preservation are evaluated on training data only or on independent test configurations.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which highlights important aspects of validation and reporting needed to strengthen the manuscript. We address each major comment below and will incorporate the requested details in the revised version.

read point-by-point responses

  1. Referee: [Abstract] Abstract (MinNEP performance paragraph): The MAE of 0.0119 eV/Ų is stated without any description of the train/validation/test partitioning, held-out terminations, or leave-one-mineral-out protocol. Because the central claim is that MinNEP reproduces the DFT surface-energy hierarchy across the full enumerated space, the absence of explicit generalization metrics on unseen terminations is load-bearing.

    Authors: We agree that the absence of explicit partitioning and generalization details in the abstract weakens the support for the central claim. In the revised manuscript we will expand both the abstract and the methods section to describe the train/validation/test split of the 764 DFT slabs, the use of held-out terminations, and any leave-one-mineral-out protocol employed to demonstrate generalization across minerals and terminations. Corresponding performance metrics on unseen data will be added. revision: yes

  2. Referee: [Abstract] Abstract and results on morphology: The claim that MinNEP 'preserves the DFT-derived morphology-determining surface-energy hierarchy' and 'reproduces the dominant Wulff-exposed facets' is presented without uncertainty quantification or error bars on the predicted energies used for ranking. This leaves open whether small extrapolation errors could alter facet ordering.

    Authors: We acknowledge that uncertainty quantification is necessary to evaluate the robustness of the reported hierarchy and facet ordering. The revised manuscript will include error estimates or confidence intervals on the MinNEP-predicted surface energies, together with a discussion of how these uncertainties affect the Wulff construction and the identification of dominant facets. revision: yes

  3. Referee: [Methods] Methods (MinNEP training description): No details are supplied on how the 764 DFT slabs were split for training versus testing, nor on whether any enumerated terminations lie outside the training distribution (different coordination or stoichiometry). This directly affects the reliability of applying the surrogate to the complete surface space.

    Authors: We agree that a complete description of the data partitioning and any extrapolation analysis is required. The revised methods section will specify the exact splitting procedure for the 764 slabs, report performance on held-out sets, and include an assessment of whether enumerated terminations with differing coordination or stoichiometry fall outside the training distribution. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper enumerates surfaces, computes DFT labels for 764 slabs, trains MinNEP on those labels, then reports MAE of the surrogate against the DFT values and applies the surrogate plus standard Wulff construction to obtain morphologies. No equation or step reduces a reported prediction to a fitted input by construction, no self-citation supplies a load-bearing uniqueness theorem, and an external XRD check is invoked for the quartz benchmark. The derivation chain therefore remains self-contained against external DFT and crystallographic data.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Based solely on abstract; ledger entries reflect standard assumptions in the domain plus the training-data scope.

axioms (2)
  • domain assumption DFT calculations supply accurate reference surface energies for training
    Invoked when MinNEP is trained to reproduce DFT values.
  • domain assumption Wulff construction yields equilibrium crystal morphologies from surface energies
    Used to link surface energies to exposed facets.

pith-pipeline@v0.9.1-grok · 5804 in / 1296 out tokens · 22746 ms · 2026-06-26T16:43:26.390361+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

10 extracted references · 10 canonical work pages

  1. [1]

    K. Li, A. Zhou, T. Liu, D. Zou, Z. Yu, Long-term performance and deterioration mechanism of novel hydrophobic coated fiber reinforced composite in marine environment, Composites, Part A 190 (2025) 108716. https://doi.org/10.1016/j.compositesa.2025.108716

    work page doi:10.1016/j.compositesa.2025.108716 2025

  2. [2]

    Sharma, R.D

    S. Sharma, R.D. Roppel, A.E. Murphy, L.W. Beegle, R. Bhartia, A. Steele, J.R. Hollis, S. Siljeström, F.M. McCubbin, S.A. Asher, W.J. Abbey, A.C. Allwood, E.L. Berger, B.L. Bleefeld, A.S. Burton, S.V. Bykov, E.L. Cardarelli, P.G. Conrad, A. Corpolongo, A.D. Czaja, L.P. DeFlores, K. Edgett, K.A. Farley, T. Fornaro, A.C. Fox, M.D. Fries, D. Harker, K. Hickma...

    work page doi:10.1038/s41586-023-06143-z 2023

  3. [3]

    W. Li, C. Xiong, Y. Zhou, Y. Zheng, J. Xing, Y. Jin, Y. Wang, Deciphering the initial hydration reaction of tricalcium aluminate based on ab initio-accurate machine learning force field, Cem. Concr. Res. 201 (2026) 108124. https://doi.org/10.1016/j.cemconres.2025.108124

    work page doi:10.1016/j.cemconres.2025.108124 2026

  4. [4]

    K. Li, A. Zhou, T. Liu, D. Zou, Z. Que, J. Ou, Low-alkalinity seawater sea sand columns reinforced with hydrophobic SFCBs and FRP hoops: a durability and energy dissipation capacity enhancement scheme for marine engineering, Eng. Struct. 350 (2026) 122059. https://doi.org/10.1016/j.engstruct.2025.122059

    work page doi:10.1016/j.engstruct.2025.122059 2026

  5. [5]

    F. Pan, Z. Liu, G. Li, Y. Zhang, Z. Fan, Y. Gang, D. Wang, Hydration behavior of quartz surfaces revealed by molecular dynamics simulations with a novel machine learning potential, J. Phys. Chem. C 129 (2025) 14829–14840. https://doi.org/10.1021/acs.jpcc.5c03759

    work page doi:10.1021/acs.jpcc.5c03759 2025

  6. [6]

    X. Xu, Z. Liu, D. Spagnoli, D.J. Armaghani, C. Qi, Adsorption mechanism of multiple water molecules on tricalcium silicate (001) surface: A DFT study, (n.d.) 1–9. https://doi.org/10.1007/s12613-024-3073-7

    work page doi:10.1007/s12613-024-3073-7

  7. [7]

    Y. Li, H. Pan, Q. Liu, X. Ming, Z. Li, Ab initio mechanism revealing for tricalcium silicate dissolution, Nat. Commun. 13 (2022) 1253. https://doi.org/10.1038/s41467-022-28932-2

    work page doi:10.1038/s41467-022-28932-2 2022

  8. [8]

    J. Yin, H. Chen, J. Qiu, W. Li, P. He, J. Li, I.A. Karimi, X. Lan, T. Wang, X. Wang, SurFF: a foundation model for surface exposure and morphology across intermetallic crystals, Nat. Comput. Sci. (2025). https://doi.org/10.1038/s43588-025-00839-0. 20

    work page doi:10.1038/s43588-025-00839-0 2025

  9. [9]

    C. Qi, X. Xu, Q. Chen, H. Liu, X. Min, A. Fourie, L. Chai, Ab initio calculation of the adsorption of As, Cd, Cr, and Hg heavy metal atoms onto the illite(001) surface: implications for soil pollution and reclamation, Environ. Pollut. 312 (2022) 120072. https://doi.org/10.1016/j.envpol.2022.120072. [10]J. Zhuo, A. Zhou, Z. Yu, T. Liu, Nanoscale toughening...

    work page doi:10.1016/j.envpol.2022.120072 2022

  10. [10]

    Zhang, Z

    P. Zhang, Z. Liu, X. Wang, R. Mi, Y. Pang, F. Pan, J. Li, Z. Wang, W. Wang, K.H. Luo, Unraveling the early-stage hydration mechanism of tricalcium silicate via machine learning potentials, Chem. Eng. J. 538 (2026) 176856. https://doi.org/10.1016/j.cej.2026.176856. [12]X. Xu, C. Qi, X.M. Aretxabaleta, C. Ma, D. Spagnoli, H. Manzano, The initial stages of c...

    work page doi:10.1016/j.cej.2026.176856 2026