Pith. sign in

REVIEW 3 major objections 5 minor 61 references

Iterative charge equilibration for fourth-generation high-dimensional neural network potentials

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Replacing a direct matrix solve with iterative minimization cuts charge equilibration scaling from cubic to quadratic in atomistic simulations.

desk verdict Useful LAMMPS/n2p2 implementation of iterative QEq with honest benchmarks, but the quadratic-scaling claim is measured at a fixed iteration cap while convergence is only demonstrated at 376 atoms. read the letter →

arxiv 2502.07907 v2 pith:XEF3YPE6 submitted 2025-02-11 cond-mat.mtrl-sci

Emir Kocer , Andreas Singraber , Jonas A. Finkler , Philipp Misof , Tsz Wai Ko , Christoph Dellago , Jörg Behler This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords chargeequilibrationfourth-generationmachinelearningpotentialshigh-dimensionalneuralnetworkiterativeminimizationBFGSlong-rangetransfermoleculardynamicsLAMMPS
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper aims to remove the bottleneck of charge equilibration in fourth-generation machine learning potentials, which describe long-range charge transfer by solving a global set of linear equations at every simulation step. The authors propose iQEq, an iterative scheme that minimizes the charge equilibration energy with a gradient-based optimizer instead of constructing and inverting the full Coulomb-plus-hardness matrix. They report that iQEq reproduces the charges and forces of the direct matrix solution while reducing the computational scaling from cubic to roughly quadratic in the number of atoms. If correct, the method makes fourth-generation potentials practical for large-scale molecular dynamics simulations of systems such as electrolytes, where local potentials fail because distant ions determine oxidation states.

What carries the argument

The load-bearing object is the charge equilibration energy $E_{\mathrm{QEq}}$, a quadratic function of the atomic partial charges whose gradient is expressed in closed form with a real-space, reciprocal-space, and self-term decomposition. Treating $E_{\mathrm{QEq}}$ as a multidimensional function lets the authors replace the $(N+1)$-dimensional matrix inversion of dQEq with a BFGS-2 minimization under the total-charge constraint, and the same gradient-based machinery is reused to obtain the Lagrange multipliers needed for efficient force evaluation. The number of iterations and the gradient tolerance become the user-controlled accuracy parameters.

What would settle it

Run iQEq on a series of increasingly large boxes of a chemically heterogeneous system (e.g., a random alloy or an ionic liquid with widely varying atomic hardness), record the number of BFGS-2 iterations to reach a fixed gradient tolerance, and compare the converged charges against a direct dQEq solve; if the iteration count grows with $N$ or the charges drift from the dQEq reference, the claimed quadratic scaling and accuracy equivalence do not generalize.

Watch

Extended reading notes

Core claim

The central claim is that the atomic charges of a fourth-generation high-dimensional neural network potential can be obtained by iteratively minimizing the charge equilibration energy $E_{\mathrm{QEq}}(Q_1,\ldots,Q_N)$ with the BFGS-2 algorithm, rather than by solving the extended linear system $A'Q'=b$ of direct charge equilibration. Because the gradient of $E_{\mathrm{QEq}}$ is available from Ewald summation without assembling the full matrix, each iteration costs $O(N\log N)$ or $O(N^2)$ depending on the real-space cutoff, and the number of iterations stays small enough that overall cost grows roughly quadratically with system size. On FeCl$_3$ in water boxes from 103 to 4512 atoms, iQEq matches dQEq charges to an RMSE of 0.0132 me after ten iterations and force components to 0.059 meV/Å, with a crossover in wall time around a few hundred atoms. The electrostatic forces are computed through the same iterative route by solving for Lagrange multipliers, avoiding the separate cubic matrix inversion of the original method.

Load-bearing premise

The quadratic scaling and the equivalence to dQEq assume that the BFGS-2 minimization converges to the same unique minimum in a roughly constant number of iterations independent of system size and chemical composition, while the benchmark covers only FeCl$_3$ in water up to 4512 atoms.

Editorial extensions

If this is right

  • Large-scale molecular dynamics with fourth-generation potentials becomes feasible: the charge equilibration step no longer dominates with cubic cost, so systems of several thousand atoms can be simulated routinely.
  • The iQEq approach is method-agnostic and can be dropped into other fourth-generation MLP frameworks that use a QEq or variational charge step, not only 4G-HDNNPs.
  • The reported convergence implies that, with a gradient tolerance of $10^{-5}$ eV/e and roughly ten iterations, iQEq and dQEq are effectively interchangeable in accuracy for aqueous FeCl$_3$.
  • Because the force calculation also avoids a second cubic matrix solve, the entire 4G-HDNNP force evaluation loop becomes approximately quadratic, shifting the bottleneck to the Ewald reciprocal-space sum.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same iterative charge minimization could be combined with a particle mesh or multilevel preconditioner to push the scaling toward quasi-linear, as the paper notes in its outlook; this would likely be a straightforward extension of the current implementation.
  • The robustness of the method probably depends on the condition number of the QEq matrix: for systems with very soft or very hard atomic species, the iteration count may grow, so a preconditioned or Krylov-based variant would be a natural extension to test.
  • The benchmark is limited to one chemistry; a stress test on a multicomponent alloy or disordered oxide with mixed oxidation states would show whether the constant-iteration assumption holds beyond FeCl$_3$ solutions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes an iterative charge equilibration (iQEq) method to replace the direct matrix inversion (dQEq) used in fourth-generation high-dimensional neural network potentials (4G-HDNNPs). The method minimizes the QEq energy with a gradient-based optimiser (BFGS-2 from GSL) under a total-charge constraint, and computes electrostatic forces using a second iterative solve for the Lagrange multipliers introduced in the force formalism. The authors implemented iQEq in LAMMPS through the n2p2 library and benchmarked it on FeCl3/water systems, reporting approximately quadratic scaling with system size (compared to cubic for dQEq), convergence of charges and forces to dQEq values for a 376-atom system (charge RMSE 0.0132 me and force RMSE 0.059 meV/Å after 10 iterations), and stable 500 ps MD trajectories.

Significance. If the claims hold, the work is useful for enabling large-scale molecular dynamics simulations with charge-transfer-capable 4G-HDNNPs, and the open-source LAMMPS implementation is a practical contribution. The underlying idea is mathematically standard: solving the same convex QEq problem by an iterative minimizer should reproduce the direct solution in the limit of convergence, and the benchmarks provide a careful check for one system. The main significance depends on two load-bearing points: that the fixed-iteration-cap timings in Fig. 3 reflect the cost of achieving dQEq-equivalent accuracy, and that the explicitly undefined iterative solve for the force Lagrange multipliers is a valid, convergent minimization. Both points need further support or clarification.

major comments (3)
  1. [Section III A and Section III B, Fig. 3 and Figs. 5-7] The quadratic scaling claim is based on wall times obtained with a fixed maximum of 15 iterations and a gradient tolerance of 10^-5 eV/e, while the accuracy against dQEq is demonstrated only for the 376-atom system. No RMSE of charges or forces with respect to dQEq is reported for the 752, 1505, 3008, or 4512 atom systems used in Fig. 3. If the number of BFGS iterations required to reach the gradient tolerance grows with system size (e.g., due to increasingly ill-conditioned QEq matrices), the measured times reflect a fixed-iteration cap rather than the cost of solving to a fixed accuracy, and the claimed quadratic scaling would not transfer to the stated accuracy. The authors should report iteration counts and charge/force RMSE versus dQEq for the larger systems, or explicitly restrict the scaling claim to a fixed number of iterations.
  2. [Section II B, Eq. (16) and the paragraph following Eq. (18)] The iterative computation of the Lagrange multipliers λ in Eq. (16) is described only as "gradient-based function minimization" without specifying the objective function. Since the bordered matrix A' in Eq. (7) is symmetric indefinite, the quadratic function whose stationary equation is A'λ = -∂E/∂Q' has no finite minimum, and a standard BFGS minimization of that quadratic is not well-posed. If the authors instead minimize a squared residual or use another reformulation, that objective must be stated explicitly; otherwise the force calculation step is not reproducible and its convergence properties cannot be assessed.
  3. [Section II B, Eqs. (9)-(10)] The constrained minimization of EQEq is described as being performed "in combination with a total charge constraint" without specifying how the constraint is enforced. It is not stated whether one charge is eliminated, a projected gradient is used, or the Lagrangian is treated as a minimizer (the latter would be a saddle-point problem, not a minimization). This is essential for the correctness and reproducibility of the iQEq charge solve, and the description should be made explicit.
minor comments (5)
  1. [Section III A] The text states that the iQEq timings use "up to 15 iterations", but it does not report how many of those iterations were actually needed for each system; without this information, the reader cannot distinguish convergence-limited from cap-limited runs.
  2. [Figures 5 and 7] The unit "me" in the charge RMSE of 0.0132 me is not defined; it presumably denotes 10^-3 elementary charge, but the abbreviation is ambiguous (conventionally "me" is the electron mass) and should be clarified, e.g., as "m e".
  3. [Eqs. (12)-(13)] The structure factor S(k) is used in Eq. (12) before it is defined in Eq. (13); the definition should be moved before its first use.
  4. [Figure 4 caption] The caption states "three different system sizes" while the text in Section III A says the speedup was tested for "four different system sizes"; the discrepancy should be resolved.
  5. [Data availability] The data availability statement says that data are available upon request, but the implementation is open-source; providing input files and scripts for the benchmark systems would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: iQEq is an independent iterative solver for the same QEq equations as dQEq; benchmarks are numerical equivalences, not fitted targets.

full rationale

The paper introduces iQEq as an iterative minimization of the same charge equilibration energy (Eq. 2, Eq. 9) whose stationarity conditions define dQEq (Eq. 6). The accuracy benchmark compares iQEq charges/forces against dQEq values computed from the same A' matrix and right-hand side for a fixed 376-atom system. This is a numerical equivalence check between two solvers of the same equations, not a prediction derived from the target: no parameter of iQEq (tolerances, iteration counts, BFGS settings) is fitted to dQEq output, and the dQEq reference is computed independently via direct linear algebra. The quadratic-scaling claim is an empirical wall-time measurement on systems up to 4512 atoms, obtained with a fixed 15-iteration cap; the lack of an analysis of iteration-count growth with system size is a robustness limitation, but not a definitional reduction. Self-citations to prior 4G-HDNNP work (Ref 37) and to Gubler et al. (Ref 51) provide context and force-formalism background, but the central iQEq algorithm and its validation are self-contained and reproducible from the equations given. No step reduces to its own input by construction, and no prediction is statistically forced by a fitted parameter.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The iQEq method rests on the standard QEq energy formalism, the Ewald summation for periodic electrostatics, the convexity of the quadratic charge equilibration energy, and the accuracy of the pre-trained 4G-HDNNP for FeCl3-water. The only user-chosen parameters are the iterative solver tolerances and iteration limits. No new physical entities are introduced.

free parameters (3)
  • Gradient tolerance = 1e-5 eV/e
    Convergence threshold for the iterative minimization of QEq energy; chosen by hand and recommended by the authors for stable trajectories.
  • Line minimization tolerance = 1e-2 eV
    Accuracy setting for the line search within the GSL minimization algorithm.
  • Maximum number of iterations = 15 (scaling) / 30 (MD)
    User-specified cap on BFGS-2 iterations; affects accuracy and performance trade-off.
assumptions (4)
  • domain assumption The QEq energy function (Eq. 2) is the correct model for charge equilibration in 4G-HDNNPs.
    Taken from prior work (Rappe and Goddard; Ko et al.) and not re-derived here.
  • domain assumption The Ewald summation (Eqs. 5 and 12) accurately represents periodic electrostatic interactions.
    Standard method, but the paper does not specify Ewald parameters (eta, cutoff) used in benchmarks.
  • standard math The QEq matrix A' is symmetric positive definite, so the quadratic energy has a unique global minimum.
    Needed for the BFGS-2 minimization to converge to the dQEq solution; not stated or analyzed in the paper.
  • domain assumption The pre-trained 4G-HDNNP for FeCl3-water (Ref 38) reproduces reference DFT charges and energies.
    The benchmark validity depends on this potential's quality, which is from the authors' prior work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Iterative charge equilibration for fourth-generation high-dimensional neural network potentials." pith.science (2026). https://pith.science/paper/XEF3YPE6

@misc{pith2026250207907,
  author       = {Pith},
  title        = {Pith review of: Iterative charge equilibration for fourth-generation high-dimensional neural network potentials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XEF3YPE6}},
  note         = {Machine review of arXiv:2502.07907}
}
abstract

Machine learning potentials (MLP) allow to perform large-scale molecular dynamics simulations with about the same accuracy as electronic structure calculations provided that the selected model is able to capture the relevant physics of the system. For systems exhibiting long-range charge transfer, fourth-generation MLPs need to be used, which take global information about the system and electrostatic interactions into account. This can be achieved in a charge equilibration (QEq) step, but the direct solution (dQEq) of the set of linear equations results in an unfavorable cubic scaling with system size making this step computationally demanding for large systems. In this work, we propose an alternative approach that is based on the iterative solution of the charge equilibration problem (iQEq) to determine the atomic partial charges. We have implemented the iQEq method, which scales quadratically with system size, in the parallel molecular dynamics software LAMMPS for the example of a fourth-generation high-dimensional neural network potential (4G-HDNNP) intended to be used in combination with the n2p2 library. The method itself is general and applicable to many different types of fourth-generation MLPs. An assessment of the accuracy and the efficiency is presented for a benchmark system of FeCl$_3$ in water.

Figures

Figures reproduced from arXiv: 2502.07907 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic structure of the direct 4G-HDNNP method for a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Schematic workflow of the 4G-HDNNP implementation [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Calculation times for the charge determination in the iQEq [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Speedup curves for three different system sizes with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Comparison of the charges [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Comparison of the charges [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Comparison of total atomic force components [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Time evolution of a) atomic charges for the Fe and Cl ions [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 21 canonical work pages

  1. [1]

    Burke ,\ title title Perspective on density functional theory , \ 10.1063/1.4704546 journal journal J

    author author K. Burke ,\ title title Perspective on density functional theory , \ 10.1063/1.4704546 journal journal J. Chem. Phys. \ volume 136 ,\ pages 150901 ( year 2012 ) NoStop

    doi:10.1063/1.4704546 2012

  2. [2]

    Behler ,\ title title Perspective: Machine learning potentials for atomistic simulations , \ 10.1063/1.4966192 journal journal J

    author author J. Behler ,\ title title Perspective: Machine learning potentials for atomistic simulations , \ 10.1063/1.4966192 journal journal J. Chem. Phys. \ volume 145 ,\ pages 170901 ( year 2016 ) NoStop

    doi:10.1063/1.4966192 2016

  3. [3]

    Behler \ and\ author G

    author author J. Behler \ and\ author G. Cs\' a nyi ,\ title title Machine learning potentials for extended systems - a perspective , \ 0.1140/epjb/s10051-021-00156-1 journal journal Eur. Phys. J. B \ volume 94 ,\ pages 142 ( year 2021 ) NoStop

    work page 2021

  4. [4]

    author author V. L. \ Deringer , author M. A. \ Caro , \ and\ author G. Cs\' a nyi ,\ title title Machine learning interatomic potentials as emerging tools for materials science , \ 10.1002/adma.201902765 journal journal Adv. Mater. \ volume 31 ,\ pages 1902765 ( year 2019 ) NoStop

    doi:10.1002/adma.201902765

  5. [5]

    No\'e , author A

    author author F. No\'e , author A. Tkatchenko , author K.-R. \ M \"u ller , \ and\ author C. Clementi ,\ title title Machine learning for molecular simulation , \ 10.1146/annurev-physchem-042018-052331 journal journal Ann. Rev. Phys. Chem. \ volume 71 ,\ pages 361--390 ( year 2020 ) NoStop

    doi:10.1146/annurev-physchem-042018-052331

  6. [6]

    u tt , author A. Tkatchenko , \ and\ author K.-R. \ M \

    author author O. T. \ Unke , author S. Chmiela , author H. E. \ Sauceda , author M. Gastegger , author I. Poltavsky , author K. T. \ Sch \"u tt , author A. Tkatchenko , \ and\ author K.-R. \ M \"u ller ,\ title title Machine learning force fields , \ 10.1021/acs.chemrev.0c01111 journal journal Chem. Rev. \ volume 121 ,\ pages 10142--10186 ( year 2021 ) NoStop

    doi:10.1021/acs.chemrev.0c01111

  7. [7]

    Friederich , author F

    author author P. Friederich , author F. H \"a se , author J. Proppe , \ and\ author A. Aspuru-Guzik ,\ title title Machine-learned potentials for next-generation matter simulations , \ 10.1038/s41563-020-0777-6 journal journal Nat. Mater \ volume 20 ,\ pages 750--761 ( year 2021 ) NoStop

    doi:10.1038/s41563-020-0777-6

  8. [8]

    Kocer , author T

    author author E. Kocer , author T. W. \ Ko , \ and\ author J. Behler ,\ title title Neural network potentials: A concise overview of methods , \ 10.1146/annurev-physchem-082720-034254 journal journal Ann. Rev. Phys. Chem. \ volume 73 ,\ pages 163--186 ( year 2022 ) NoStop

    doi:10.1146/annurev-physchem-082720-034254

Show all 61 references
  1. [9]

    author author T. B. \ Blank , author S. D. \ Brown , author A. W. \ Calhoun , \ and\ author D. J. \ Doren ,\ title title Neural network models of potential energy surfaces , \ 10.1063/1.469597 journal journal J. Chem. Phys. \ volume 103 ,\ pages 4129--4137 ( year 1995 ) NoStop

    doi

  2. [10]

    author author T. W. \ Ko , author J. A. \ Finkler , author S. Goedecker , \ and\ author J. Behler ,\ title title General-purpose machine learning potentials capturing nonlocal charge transfer , \ 10.1021/acs.accounts.0c00689 journal journal Acc. Chem. Res. \ volume 54 ,\ pages...

    doi

  3. [11]

    Lorenz , author A

    author author S. Lorenz , author A. Gro , \ and\ author M. Scheffler ,\ title title Representing high-dimensional potential-energy surfaces for reactions at surfaces by neural networks , \ 10.1016/j.cplett.2004.07.076 journal journal Chem. Phys. Lett. \ volume 395 ,\ pages 210...

    2004 doi

  4. [12]

    Behler , author S

    author author J. Behler , author S. Lorenz , \ and\ author K. Reuter ,\ title title Representing molecule-surface interactions with symmetry-adapted neural networks , \ 10.1063/1.2746232 journal journal J. Chem. Phys. \ volume 127 ,\ pages 014705 ( year 2007 ) NoStop

    doi

  5. [13]

    Gassner , author M

    author author H. Gassner , author M. Probst , author A. Lauenstein , \ and\ author K. Hermansson ,\ title title Representation of intermolecular potential functions by neural networks , \ 10.1021/jp972209d journal journal J. Phys. Chem. A \ volume 102 ,\ pages 4596--4605 ( yea...

    doi

  6. [14]

    Behler \ and\ author M

    author author J. Behler \ and\ author M. Parrinello ,\ title title Generalized neural-network representation of high-dimensional potential-energy surfaces , \ 10.1103/PhysRevLett.98.146401 journal journal Phys. Rev. Lett. \ volume 98 ,\ pages 146401 ( year 2007 ) NoStop

    doi

  7. [15]

    author author A. P. \ Bart\' o k , author M. C. \ Payne , author R. Kondor , \ and\ author G. Cs\' a nyi ,\ title title Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons , \ 10.1103/PhysRevLett.104.136403 journal journal Phys. Rev. Let...

    doi

  8. [16]

    author author A. P. \ Thompson , author L. P. \ Swiler , author C. R. \ Trott , author S. M. \ Foiles , \ and\ author G. J. \ Tucker ,\ title title Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials , \ 10.1016/j.jcp.2014.12.0...

    doi

  9. [17]

    author author A. V. \ Shapeev ,\ title title Moment tensor potentials: a class of systematically improvable interatomic potentials , \ 10.1137/15M1054183 journal journal Multiscale Model. Simul. \ volume 14 ,\ pages 1153--1173 ( year 2016 ) NoStop

    doi

  10. [18]

    Drautz ,\ title title Atomic cluster expansion for accurate and transferable interatomic potentials , \ 10.1103/PhysRevB.99.014104 journal journal Phys

    author author R. Drautz ,\ title title Atomic cluster expansion for accurate and transferable interatomic potentials , \ 10.1103/PhysRevB.99.014104 journal journal Phys. Rev. B \ volume 99 ,\ pages 014104 ( year 2019 ) NoStop

    2019 doi

  11. [19]

    author author J. S. \ Smith , author O. Isayev , \ and\ author A. E. \ Roitberg ,\ title title ANI -1: An extensible neural network potential with dft accuracy at force field computational cost , \ 10.1039/C6SC05720A journal journal Chem. Sci. \ volume 8 ,\ pages 3192--3203 ( ...

    doi

  12. [20]

    Liu \ and\ author J

    author author M. Liu \ and\ author J. R. \ Kitchin ,\ title title SingleNN : A modified Behler-Parrinello neural network with shared weights for atomistic simulations with transferability , \ 10.1021/acs.jpcc.0c04225 journal journal J. Phys. Chem. C \ volume 124 ,\ pages 17811...

    doi

  13. [21]

    Gilmer , author S

    author author J. Gilmer , author S. S. \ Schoenholz , author P. F. \ Riley , author O. Vinyals , \ and\ author G. E. \ Dahl ,\ title title Neural message passing for quantum chemistry , \ in\ http://proceedings.mlr.press/v70/gilmer17a.html booktitle Proceedings of the 34th Int...

    2017

  14. [22]

    u tt , author H. E. \ Sauceda , author P.-J. \ Kindermans , author A. Tkatchenko , \ and\ author K.-R. \ M \

    author author K. T. \ Sch \"u tt , author H. E. \ Sauceda , author P.-J. \ Kindermans , author A. Tkatchenko , \ and\ author K.-R. \ M \"u ller ,\ title title SchNet - a deep learning architecture for molecules and materials , \ 10.1063/1.5019779 journal journal J. Chem. Phys....

    doi

  15. [23]

    Zubatyuk , author J

    author author R. Zubatyuk , author J. S. \ Smith , author J. Leszczynski , \ and\ author O. Isayev ,\ title title Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network , \ 10.1126/sciadv.aav6490 journal journal Science ...

    doi

  16. [24]

    Batzner , author A

    author author S. Batzner , author A. Musaelian , author L. Sun , author M. Geiger , author J. P. \ Mailoa , author M. Kornbluth , author N. Molinari , author T. E. \ Smidt , \ and\ author B. Kozinsky ,\ title title SE(3) -equivariant graph neural networks for data-efficient an...

    doi

  17. [25]

    Batatia , author D

    author author I. Batatia , author D. P. \ Kovacs , author G. Simm , author C. Ortner , \ and\ author G. Csanyi ,\ title title MACE : Higher order equivariant message passing neural networks for fast and accurate force fields , \ in\ 10.48550/arXiv.2206.07697 booktitle Advances...

    doi arXiv

  18. [26]

    Houlding , author S

    author author S. Houlding , author S. Y. \ Liem , \ and\ author P. L. A. \ Popelier ,\ title title A polarizable high-rank quantum topological electrostatic potential developed using neural networks: Molecular dynamics simulations on the hydrogen fluoride dimer , \ 10.1002/qua...

    doi

  19. [27]

    Artrith , author T

    author author N. Artrith , author T. Morawietz , \ and\ author J. Behler ,\ title title High-dimensional neural-network potentials for multicomponent systems: Applications to zinc oxide , \ 10.1103/PhysRevB.83.153101 journal journal Phys. Rev. B \ volume 83 ,\ pages 153101 ( y...

    doi

  20. [28]

    Morawietz , author V

    author author T. Morawietz , author V. Sharma , \ and\ author J. Behler ,\ title title A neural network potential-energy surface for the water dimer based on environment-dependent atomic energies and charges , \ 10.1063/1.3682557 journal journal J. Chem. Phys. \ volume 136 ,\ ...

    doi

  21. [29]

    author author O. T. \ Unke \ and\ author M. Meuwly ,\ title title PhysNet : A neural network for predicting energies, forces, dipole moments, and partial charges , \ 10.1021/acs.jctc.9b00181 journal journal J. Chem. Theory Comput. \ volume 15 ,\ pages 3678--3693 ( year 2019 ) NoStop

    doi

  22. [30]

    Yao , author J

    author author K. Yao , author J. E. \ Herr , author D. W. \ Toth , author R. Mckintyre , \ and\ author J. Parkhill ,\ title title The tensormol-0.1 model chemistry: a neural network augmented with long-range physics , \ 10.1039/C7SC04934J journal journal Chem. Sci. \ volume 9 ...

    doi

  23. [31]

    Grisafi \ and\ author M

    author author A. Grisafi \ and\ author M. Ceriotti ,\ title title Incorporating long-range physics in atomic-scale machine learning , \ 10.1063/1.5128375@jcp.2020.MACH2020.issue-1 journal journal J. Chem. Phys. \ volume 151 ,\ pages 204105 ( year 2019 ) NoStop

    2020 doi

  24. [32]

    Gao \ and\ author R

    author author A. Gao \ and\ author R. C. \ Remsing ,\ title title Self-consistent determination of long-range electrostatics in neural network potentials , \ @noop journal journal Nature Comm. \ volume 13 ,\ pages 1572 ( year 2022 ) NoStop

    2022

  25. [33]

    Zhang , author H

    author author L. Zhang , author H. Wang , author M. C. \ Muniz , author A. Z. \ Panagiotopoulos , author R. Car , \ and\ author W. E ,\ title title A deep potential model with long-range electrostatic interactions , \ @noop journal journal J. Chem. Phys. \ volume 156 ,\ pages ...

    2022

  26. [34]

    Yue , author M

    author author S. Yue , author M. C. \ Muniz , author M. F. C. \ Andrade , author L. Zhang , author R. Car , \ and\ author A. Z. \ Panagiotopoulos ,\ title title When do short-range atomistic machine-learning models fall short? \ 10.1063/5.0031215 journal journal J. Chem. Phys....

    doi

  27. [35]

    author author P. P. \ Ewald ,\ title title Die Berechnung optischer und elektrostatischer Gitterpotentiale , \ 10.1002/andp.19213690304 journal journal Ann. Phys. \ volume 64 ,\ pages 253--287 ( year 1921 ) NoStop

    1921 doi

  28. [36]

    author author A. Y. \ Toukmaji \ and\ author J. A. B. Board ,\ title title Ewald summation techniques in perspective: a survey , \ 10.1016/0010-4655(96)00016-1 journal journal Comp. Phys. Comm. \ volume 95 ,\ pages 73--92 ( year 1996 ) NoStop

    doi

  29. [37]

    author author T. W. \ Ko , author J. A. \ Finkler , author S. Goedecker , \ and\ author J. Behler ,\ title title A fourth-generation high-dimensional neural network potential with accurate electrostatics including non-local charge transfer , \ 10.1038/s41467-020-20427-2 journa...

    doi

  30. [38]

    Kocer , author R

    author author E. Kocer , author R. El Haouari , author C. Dellago , \ and\ author J. Behler ,\ title title Machine learning potentials for redox chemistry in solution , \ @noop journal journal arXiv preprint arXiv:2410.03299 \ ( year 2024 ) NoStop

    arXiv

  31. [39]

    author author S. A. \ Ghasemi , author A. Hofstetter , author S. Saha , \ and\ author S. Goedecker ,\ title title Interatomic potentials for ionic systems with density functional accuracy based on charge densities obtained by a neural network , \ 10.1103/PhysRevB.92.045131 jou...

    doi

  32. [40]

    author author A. K. \ Rappe \ and\ author W. A. \ Goddard , III ,\ title title Charge equilibration for molecular dynamics simulations , \ 10.1021/j100161a070 journal journal J. Phys. Chem. \ volume 95 ,\ pages 3358--3363 ( year 1991 ) NoStop

    doi

  33. [41]

    Faraji , author S

    author author S. Faraji , author S. A. \ Ghasemi , author S. Rostami , author R. Rasoulkhani , author B. Schaefer , author S. Goedecker , \ and\ author M. Amsler ,\ title title High accuracy and transferability of a neural network potential through charge equilibration for cal...

    doi

  34. [42]

    Faraji , author S

    author author S. Faraji , author S. A. \ Ghasemi , author B. Parsaeifard , \ and\ author S. Goedecker ,\ title title Surface reconstructions and premelting of the (100) CaF _2 surface , \ 10.1039/C9CP02213A journal journal Phys. Chem. Chem. Phys. \ volume 21 ,\ pages 16270--16...

    doi

  35. [43]

    author author H. A. \ Eivari , author S. A. \ Ghasemi , author H. Tahmasbi , author S. Rostami , author S. Faraji , author R. Rasoulkhani , author S. Goedecker , \ and\ author M. Amsler ,\ title title Two-dimensional hexagonal sheet of TiO _2 , \ 10.1021/acs.chemmater.7b02031 ...

    doi

  36. [44]

    author author C. G. \ Staacke , author S. Wengert , author C. Kunkel , author G. Csányi , author K. Reuter , \ and\ author J. T. \ Margraf ,\ title title Kernel charge equilibration: efficient and accurate prediction of molecular dipole moments with a machine-learning enhanced...

    doi

  37. [45]

    Jacobson , author J

    author author L. Jacobson , author J. Stevenson , author F. Ramezanghorbani , author D. Ghoreishi , author K. Leswing , author E. Harder , \ and\ author R. Abel ,\ title title Transferable neural network potential energy surfaces for closed-shell organic molecules: Extension t...

    doi

  38. [46]

    author author E. R. \ Khajehpasha , author J. A. \ Finkler , author T. D. \ K \"u hne , \ and\ author S. A. \ Ghasemi ,\ title title CENT2 : Improved charge equilibration via neural network technique , \ 10.1103/PhysRevB.105.144106 journal journal Phys. Rev. B \ volume 105 ,\ ...

    doi

  39. [47]

    author author T. W. \ Ko , author J. A. \ Finkler , author S. Goedecker , \ and\ author J. Behler ,\ title title Accurate fourth-generation machine learning potentials by electrostatic embedding , \ 10.1021/acs.jctc.2c01146 journal journal J. Chem. Theory Comput. \ volume 19 ,...

    doi

  40. [48]

    Xie , author K

    author author X. Xie , author K. A. \ Persson , \ and\ author D. W. \ Small ,\ title title Incorporating electronic information into machine learning potential energy surfaces via approaching the ground-state electronic energy as a function of atom-based electronic populations...

    doi

  41. [49]

    Shaidu , author F

    author author Y. Shaidu , author F. Pellegrini , author E. K \"u c \"u kbenli , author R. Lot , \ and\ author S. de Gironcoli ,\ title title Incorporating long-range electrostatics in neural network potentials via variational charge equilibration from shortsighted ingredients ...

    doi

  42. [50]

    Behler ,\ title title Atom-centered symmetry functions for constructing high-dimensional neural network potentials , \ 10.1063/1.3553717 journal journal J

    author author J. Behler ,\ title title Atom-centered symmetry functions for constructing high-dimensional neural network potentials , \ 10.1063/1.3553717 journal journal J. Chem. Phys. \ volume 134 ,\ pages 074106 ( year 2011 ) NoStop

    doi

  43. [51]

    Gubler , author J

    author author M. Gubler , author J. A. \ Finkler , author M. R. \ Sch\"afer , author J. Behler , \ and\ author S. Goedecker ,\ title title Accelerating fourth-generation machine learning potentials by quasi-linear scaling particle mesh charge equilibration , \ 10.1021/acs.jctc...

    doi

  44. [52]

    Plimpton ,\ title title Fast parallel algorithms for short-range molecular dynamics , \ 10.2172/10176421 journal journal J

    author author S. Plimpton ,\ title title Fast parallel algorithms for short-range molecular dynamics , \ 10.2172/10176421 journal journal J. Comp. Phys. \ volume 117 ,\ pages 1 ( year 1995 ) NoStop

    1995 doi

  45. [53]

    Singraber , author J

    author author A. Singraber , author J. Behler , \ and\ author C. Dellago ,\ title title Library-based LAMMPS implementation of high-dimensional neural network potentials , \ 10.1021/acs.jctc.8b00770 journal journal J. Chem. Theory Comput. \ volume 15 ,\ pages 1827--1840 ( year...

    doi

  46. [54]

    Behler ,\ title title Four generations of high-dimensional neural network potentials , \ 10.1021/acs.chemrev.0c00868 journal journal Chem

    author author J. Behler ,\ title title Four generations of high-dimensional neural network potentials , \ 10.1021/acs.chemrev.0c00868 journal journal Chem. Rev. \ volume 121 ,\ pages 10037--10072 ( year 2021 ) NoStop

    2021 doi

  47. [55]

    Verstraelen , author V

    author author T. Verstraelen , author V. V. \ Speybroeck , \ and\ author M. Waroquier ,\ title title The electronegativity equalization method and the split charge equilibration applied to organic systems: Parametrization, validation, and comparison , \ 10.1063/1.3187034 journ...

    doi

  48. [56]

    author author C. E. \ Wilmer , author K. C. \ Kim , \ and\ author R. Q. \ Snurr ,\ title title An extended charge equilibration method , \ 10.1021/jz3008485 journal journal J. Phys. Chem. Lett. \ volume 3 ,\ pages 2506--2511 ( year 2012 ) NoStop

    doi

  49. [57]

    author author F. L. \ Hirshfeld ,\ title title Bonded-atom fragments for describing molecular charge densities , \ 10.1007/BF00549096 journal journal Theor. Chim. Acta \ volume 44 ,\ pages 129--138 ( year 1977 ) NoStop

    1977 doi

  50. [58]

    Nakano ,\ title title Parallel multilevel preconditioned conjugate-gradient approach to variable-charge molecular dynamics , \ 10.1016/S0010-4655(97)00041-6 journal journal Comp

    author author A. Nakano ,\ title title Parallel multilevel preconditioned conjugate-gradient approach to variable-charge molecular dynamics , \ 10.1016/S0010-4655(97)00041-6 journal journal Comp. Phys. Comm. \ volume 104 ,\ pages 59--69 ( year 1997 ) NoStop

    doi

  51. [59]

    Galassi , author J

    author author M. Galassi , author J. Davies , author J. Theiler , author B. Gough , author G. Jungman , author P. Alken , author M. Booth , author F. Rossi , \ and\ author R. Ulerich ,\ @noop title GNU scientific library \ ( publisher Network Theory Limited Godalming ,\ year 2...

    2002

  52. [60]

    Fletcher ,\ @noop title Practical methods of optimization \ ( publisher John Wiley & Sons ,\ year 2000 ) NoStop

    author author R. Fletcher ,\ @noop title Practical methods of optimization \ ( publisher John Wiley & Sons ,\ year 2000 ) NoStop

    2000

  53. [61]

    author author Message Passing Interface Forum ,\ https://www.mpi-forum.org/docs/mpi-4.0/mpi40-report.pdf title MPI : A Message-Passing Interface Standard Version 4.0 ( year 2021 ) NoStop

    2021

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.