pith. sign in

arxiv: 2605.28798 · v1 · pith:5PCKQEILnew · submitted 2026-05-27 · ⚛️ physics.chem-ph · cond-mat.mtrl-sci· physics.comp-ph

How reproducible are first-principles simulations of liquid water?

Niamh ONeill , Benjamin X. Shi , William J. Baldwin , Albert P. Bartok , Chris J. Pickard , Angelos Michaelides , Gabor Csanyi , Timothy C. Berkelbach This is my paper

Pith reviewed 2026-06-29 09:16 UTC · model grok-4.3

classification ⚛️ physics.chem-ph cond-mat.mtrl-sciphysics.comp-ph
keywords liquid waterrevPBE-D3ab initio molecular dynamicsreproducibilitymachine learning potentialsdensitydiffusion coefficientbasis set convergence
0
0 comments X

The pith

Previous revPBE-D3 simulations of liquid water overestimated density and underestimated diffusion due to basis set incompleteness and pseudopotential inconsistencies.

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

The paper reveals that different studies of liquid water using the same revPBE-D3 density functional produced results differing by over 20 percent in the diffusion coefficient and 10 percent in the density. These inconsistencies stem from incomplete basis sets, mismatched pseudopotentials, and limited statistical sampling in the molecular dynamics runs. The authors train machine-learning interatomic potentials on carefully converged DFT data to enable long, well-sampled simulations that reach agreement across six independent codes. The new consensus values correct the earlier overestimation of density and underestimation of diffusion, supplying a reliable reference for checking DFT-based simulations of water.

Core claim

Previous studies of liquid water using the same widely-used density functional (revPBE-D3) exhibit significant discrepancies with one another, varying by over 20% in the diffusion coefficient and 10% in the density. By combining modern long-range machine-learning interatomic potentials that enable robust statistical sampling with carefully converged DFT training data, we resolve these discrepancies, achieving consensus across six diverse community codes. Our predictions differ markedly from previous literature: most previous results overestimate the density and underestimate the diffusion coefficient of revPBE-D3 water due to basis set incompleteness and pseudopotential inconsistencies, coup

What carries the argument

Long-range machine-learning interatomic potentials trained on converged DFT data, which reproduce the revPBE-D3 potential energy surface while permitting extensive sampling of liquid-phase properties.

If this is right

  • Consensus values for the density and diffusion coefficient of revPBE-D3 liquid water now exist as benchmarks for other implementations.
  • Most earlier literature values for these properties must be viewed as unreliable due to technical setup errors.
  • Agreement across six independent community codes confirms the corrected results.
  • Reliable reference data is now available for systematic tests of new density functionals and numerical approximations.

Where Pith is reading between the lines

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

  • Reproducibility issues of this type may affect DFT simulations of other molecular liquids.
  • Machine-learning potentials could be used more broadly to reach statistical convergence in ab initio simulations of condensed phases.
  • Future studies should routinely verify basis set completeness and pseudopotential consistency before reporting liquid properties.

Load-bearing premise

The machine-learning interatomic potentials trained on the chosen DFT data faithfully reproduce the converged revPBE-D3 potential energy surface without introducing systematic biases in the liquid-phase properties being benchmarked.

What would settle it

Direct ab initio molecular dynamics runs using fully converged basis sets and consistent pseudopotentials across multiple codes, without machine-learning potentials, that either match or fail to match the new benchmark density and diffusion values.

Figures

Figures reproduced from arXiv: 2605.28798 by Albert P. Bartok, Angelos Michaelides, Benjamin X. Shi, Chris J. Pickard, Gabor Csanyi, Niamh ONeill, Timothy C. Berkelbach, William J. Baldwin.

Figure 1
Figure 1. Figure 1: Reproducibility of revPBE-D3 for density and diffusion coefficient of water at ambient conditions. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p026_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗
read the original abstract

Liquid water is fundamentally important, and its accurate computer simulation has been the driving force for myriad methodological developments. Ab initio molecular dynamics with forces obtained from density functional theory (DFT) is now a standard tool widely used by researchers. However, we reveal that previous studies of liquid water using the same widely-used density functional (revPBE-D3) exhibit significant discrepancies with one another, varying by over 20% in the diffusion coefficient and 10% in the density, raising fundamental questions about reproducibility. By combining modern long-range machine-learning interatomic potentials that enable robust statistical sampling with carefully converged DFT training data, we resolve these discrepancies, achieving consensus across six diverse community codes. Our predictions differ markedly from previous literature: we show that most previous results overestimate the density and underestimate the diffusion coefficient of revPBE-D3 water due to basis set incompleteness and pseudopotential inconsistencies, coupled with limitations in statistical sampling (in some cases). These benchmark values provide a reliable reference for validating current and future implementations of DFT-based ab initio molecular dynamics. Reaching agreement establishes confidence and credibility and serves as a prerequisite for the systematic assessment of new density functionals and numerical approximations.

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

2 major / 0 minor

Summary. The manuscript examines reproducibility in ab initio molecular dynamics simulations of liquid water with the revPBE-D3 functional. It identifies discrepancies exceeding 20% in diffusion coefficients and 10% in densities across prior studies, attributes these to basis-set incompleteness, pseudopotential inconsistencies, and insufficient statistical sampling, and claims to resolve them via machine-learning interatomic potentials trained on carefully converged DFT data from six independent codes, yielding consensus benchmark values that differ from previous literature and serve as reliable references for future DFT-based AIMD work.

Significance. If the central claim holds, the work supplies important consensus benchmarks for revPBE-D3 liquid water and clarifies sources of variability in community AIMD implementations, which would be useful for validating codes and assessing new functionals. The approach of combining converged multi-code DFT training data with ML potentials for extended sampling is a methodological strength that could improve reproducibility in the field.

major comments (2)
  1. [Abstract] Abstract: The claim that ML potentials trained on converged DFT data 'resolve these discrepancies' and 'achieve consensus' rests on the unshown assumption that the learned potentials introduce no systematic bias in liquid-phase density or diffusion relative to direct DFT; however, the manuscript supplies no force or energy error statistics on liquid configurations, no direct DFT-vs-ML comparisons of RDFs or transport coefficients, and no cross-validation against the six codes' raw DFT trajectories.
  2. [Abstract] Abstract: The statement that most previous results 'overestimate the density and underestimate the diffusion coefficient' due to basis set incompleteness and pseudopotential inconsistencies lacks accompanying quantitative details on training-set convergence, error bars on the ML potentials, or statistical uncertainties on the final benchmarks, so the central claim that discrepancies are resolved cannot be evaluated from the provided evidence.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for highlighting areas where the abstract could more explicitly support our central claims. We address each comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that ML potentials trained on converged DFT data 'resolve these discrepancies' and 'achieve consensus' rests on the unshown assumption that the learned potentials introduce no systematic bias in liquid-phase density or diffusion relative to direct DFT; however, the manuscript supplies no force or energy error statistics on liquid configurations, no direct DFT-vs-ML comparisons of RDFs or transport coefficients, and no cross-validation against the six codes' raw DFT trajectories.

    Authors: We agree that explicit validation metrics are required to substantiate the absence of systematic bias. The main text presents training procedures and overall agreement across codes, but does not include the specific liquid-configuration error statistics, RDF/transport comparisons, or per-code cross-validation requested. We will add these quantitative results (force/energy RMSE on held-out liquid snapshots, direct DFT-vs-ML RDF and diffusion comparisons, and code-by-code trajectory validation) to both the abstract and the results section. revision: yes

  2. Referee: [Abstract] Abstract: The statement that most previous results 'overestimate the density and underestimate the diffusion coefficient' due to basis set incompleteness and pseudopotential inconsistencies lacks accompanying quantitative details on training-set convergence, error bars on the ML potentials, or statistical uncertainties on the final benchmarks, so the central claim that discrepancies are resolved cannot be evaluated from the provided evidence.

    Authors: We acknowledge that the abstract currently omits the requested quantitative details. While the body of the manuscript reports training-set sizes and convergence tests, it does not tabulate the precise error bars on the ML potentials or the statistical uncertainties on the final consensus density and diffusion values. In the revision we will insert these numbers (training-set convergence metrics, ML potential error bars, and bootstrap uncertainties on the benchmarks) into the abstract and a dedicated methods/results subsection. revision: yes

Circularity Check

0 steps flagged

No significant circularity; benchmarks obtained via direct multi-code comparison on converged DFT data

full rationale

The derivation relies on training ML potentials to converged DFT reference calculations (energies/forces) generated independently across six codes, followed by long-timescale sampling to extract liquid observables. These observables are not inputs to the training procedure and do not reduce by construction to any fitted parameter or self-citation. The abstract and reader's summary indicate the reported consensus values arise from explicit cross-code validation rather than tautological re-expression of the training data. No self-definitional, fitted-input-renamed-as-prediction, or load-bearing self-citation steps are present.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger reflects the methodology described therein; the central claim depends on the fidelity of ML potentials to converged DFT and on the assumption that the chosen training data capture the relevant liquid configurations.

free parameters (1)
  • ML interatomic potential parameters
    Parameters of the machine-learning models are fitted to the DFT training data to enable extended sampling.
axioms (1)
  • domain assumption revPBE-D3 is a suitable and representative density functional for liquid water benchmarking
    The entire study is performed with this specific functional as the target.

pith-pipeline@v0.9.1-grok · 5773 in / 1230 out tokens · 38517 ms · 2026-06-29T09:16:45.799379+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

124 extracted references · 8 canonical work pages · 2 internal anchors

  1. [1]

    Lejaeghere, G

    K. Lejaeghere, G. Bihlmayer, T. Bj¨ orkman, P. Blaha, S. Bl¨ ugel, V. Blum, D. Caliste, I. E. Castelli, S. J. Clark, A. Dal Corso, S. de Gironcoli, T. Deutsch, J. K. Dewhurst, I. Di Marco, C. Draxl, M. Du lak, O. Eriksson, J. A. Flores- Livas, K. F. Garrity, L. Genovese, P. Giannozzi, M. Gi- antomassi, S. Goedecker, X. Gonze, O. Gr˚ an¨ as, E. K. U. Gross...

    2016

  2. [2]

    Bosoni, L

    E. Bosoni, L. Beal, M. Bercx, P. Blaha, S. Bl¨ ugel, J. Br¨ oder, M. Callsen, S. Cottenier, A. Degomme, V. Dikan, K. Eimre, E. Flage-Larsen, M. Fornari, A. Garcia, L. Genovese, M. Giantomassi, S. P. Huber, H. Janssen, G. Kastlunger, M. Krack, G. Kresse, T. D. K¨ uhne, K. Lejaeghere, G. K. H. Madsen, M. Marsman, N. Marzari, G. Michalicek, H. Mirhosseini, T...

    2024

  3. [3]

    M. J. van Setten, F. Caruso, S. Sharifzadeh, X. Ren, M. Scheffler, F. Liu, J. Lischner, L. Lin, J. R. Deslippe, S. G. Louie, C. Yang, F. Weigend, J. B. Neaton, F. Evers, and P. Rinke, GW100: Benchmarking G0W0 for molecular systems, J. Chem. Theory Comput.11, 5665 (2015)

    2015

  4. [4]

    Della Pia, B

    F. Della Pia, B. X. Shi, Y. S. Al-Hamdani, D. Alf´ e, T. A. Anderson, M. Barborini, A. Benali, M. Casula, N. D. Drum- mond, M. Dubeck´ y, C. Filippi, P. R. C. Kent, J. T. Kro- gel, P. L´ opez R ´ ıos, A. L¨ uchow, Y. Luo, A. Michaelides, 6 L. Mitas, K. Nakano, R. J. Needs, M. C. Per, A. Sce- mama, J. Schultze, R. Shinde, E. Slootman, S. Sorella, A. Tkatch...

    2025

  5. [5]

    Yeh and G

    I.-C. Yeh and G. Hummer, System-size dependence of dif- fusion coefficients and viscosities from molecular dynam- ics simulations with periodic boundary conditions, J. Phys. Chem. B108, 15873 (2004)

    2004

  6. [6]

    L. R. Pestana, N. Mardirossian, M. Head-Gordon, and T. Head-Gordon, Ab initio molecular dynamics simulations of liquid water using high quality meta-GGA functionals, Chem. Sci.8, 3554 (2017)

    2017

  7. [7]

    Galib, T

    M. Galib, T. T. Duignan, Y. Misteli, M. D. Baer, G. K. Schenter, J. Hutter, and C. J. Mundy, Mass density fluctu- ations in quantum and classical descriptions of liquid water, J. Chem. Phys.146, 244501 (2017)

    2017

  8. [8]

    in-and-out

    S. G. H. Brookes, V. Kapil, A. Michaelides, and C. Schran, CO2 hydration at the air–water interface: A surface- mediated “in-and-out” mechanism, Proc. Natl. Acad. Sci. U. S. A.122, e2502684122 (2025)

    2025

  9. [9]

    Ferretti, G

    A. Ferretti, G. Melani, L. Benedetti, R. A. Sorodoc, A. For- tunelli, and G. Brancato, Accurate simulations of water and aqueous solutions through fine-tuned dispersion-corrected density functional theory and machine-learning interatomic potentials, J. Chem. Inf. Model.65, 12437 (2025)

    2025

  10. [10]

    Montero de Hijes, C

    P. Montero de Hijes, C. Dellago, R. Jinnouchi, and G. Kresse, Density isobar of water and melting tempera- ture of ice: Assessing common density functionals, J. Chem. Phys.161, 131102 (2024)

    2024

  11. [13]

    N. V. S. Avula, M. L. Klein, and S. Balasubramanian, Understanding the anomalous diffusion of water in aque- ous electrolytes using machine learned potentials, J. Phys. Chem. Lett.14, 9500 (2023)

    2023

  12. [16]

    Marsalek and T

    O. Marsalek and T. E. Markland, Quantum dynamics and spectroscopy ofab initioliquid water: The interplay of nu- clear and electronic quantum effects, J. Phys. Chem. Lett. 8, 1545 (2017)

    2017

  13. [20]

    M. J. Gillan, D. Alf` e, and A. Michaelides, Perspective: How good is DFT for water?, J. Chem. Phys.144, 130901 (2016)

    2016

  14. [21]

    M. E. Tuckerman, K. Laasonen, M. Sprik, and M. Parrinello, Ab initio simulations of water and water ions, J. Phys.: Con- dens. Matter6, A93 (1994)

    1994

  15. [22]

    M. V. Fern´ andez-Serra and E. Artacho, Network equilibra- tion and first-principles liquid water, J. Chem. Phys.121, 11136 (2004)

    2004

  16. [23]

    J. C. Grossman, E. Schwegler, E. W. Draeger, F. Gygi, and G. Galli, Towards an assessment of the accuracy of density functional theory for first principles simulations of water, J. Chem. Phys.120, 300 (2004)

    2004

  17. [24]

    Schwegler, J

    E. Schwegler, J. C. Grossman, F. Gygi, and G. Galli, To- wards an assessment of the accuracy of density functional theory for first principles simulations of water. II, J. Chem. Phys.121, 5400 (2004)

    2004

  18. [25]

    VandeVondele, F

    J. VandeVondele, F. Mohamed, M. Krack, J. Hutter, M. Sprik, and M. Parrinello, The influence of temperature and density functional models in ab initio molecular dynam- ics simulation of liquid water, J. Chem. Phys.122, 014515 (2004)

    2004

  19. [26]

    P. H.-L. Sit and N. Marzari, Static and dynamical properties of heavy water at ambient conditions from first-principles molecular dynamics, J. Chem. Phys.122, 204510 (2005)

    2005

  20. [27]

    Santra, A

    B. Santra, A. Michaelides, and M. Scheffler, Coupled clus- ter benchmarks of water monomers and dimers extracted from density-functional theory liquid water: The importance of monomer deformations, J. Chem. Phys.131, 124509 (2009)

    2009

  21. [28]

    Ceriotti, W

    M. Ceriotti, W. Fang, P. G. Kusalik, R. H. McKenzie, A. Michaelides, M. A. Morales, and T. E. Markland, Nu- clear quantum effects in water and aqueous systems: Ex- periment, theory, and current challenges, Chem. Rev.116, 7529 (2016)

    2016

  22. [29]

    Chen, H.-Y

    M. Chen, H.-Y. Ko, R. C. Remsing, M. F. Calegari Andrade, B. Santra, Z. Sun, A. Selloni, R. Car, M. L. Klein, J. P. Perdew, and X. Wu, Ab initio theory and modeling of water, Proc. Natl. Acad. Sci. U. S. A.114, 10846 (2017)

    2017

  23. [30]

    M. D. LaCount and F. Gygi, Ensemble first-principles molecular dynamics simulations of water using the SCAN meta-GGA density functional, J. Chem. Phys.151, 164101 (2019)

    2019

  24. [31]

    Zhang, F

    C. Zhang, F. Tang, M. Chen, J. Xu, L. Zhang, D. Y. Qiu, J. P. Perdew, M. L. Klein, and X. Wu, Modeling Liquid Water by Climbing up Jacob’s Ladder in Density Functional Theory Facilitated by Using Deep Neural Network Poten- tials, J. Phys. Chem. B125, 11444 (2021)

    2021

  25. [32]

    Villard, M

    J. Villard, M. P. Bircher, and U. Rothlisberger, Structure and dynamics of liquid water from ab initio simulations: Adding Minnesota density functionals to Jacob’s ladder, Chem. Sci.15, 4434 (2024)

    2024

  26. [33]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gra- dient approximation made simple, Phys. Rev. Lett.77, 3865 (1996)

    1996

  27. [34]

    Generalized gradi- ent approximation made simple

    Y. Zhang and W. Yang, Comment on “Generalized gradi- ent approximation made simple”, Phys. Rev. Lett.80, 890 (1998)

    1998

  28. [35]

    Grimme, J

    S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu, J. Chem. Phys.132, 154104 (2010). 7

    2010

  29. [36]

    Cheng, E

    B. Cheng, E. A. Engel, J. Behler, C. Dellago, and M. Ce- riotti, Ab initio thermodynamics of liquid and solid water, Proc. Natl. Acad. Sci. U. S. A.116, 1110 (2019)

    2019

  30. [37]

    first principles

    S. L. Bore and F. Paesani, Realistic phase diagram of wa- ter from “first principles” data-driven quantum simulations, Nat. Commun.14, 3349 (2023)

    2023

  31. [38]

    N. H. D` ao, D. S. King, F. Stein, X. Wang, D. Kim, G. Bran- denburg, J. Hutter, and B. Cheng, Systematic trends in water properties across Jacob’s ladder density functionals, ChemRxiv 10.26434/chemrxiv.15000644/v1 (2026)

    work page doi:10.26434/chemrxiv.15000644/v1 2026

  32. [39]

    More converged, less accurate? Reassessing standard choices for ab initio water using machine learning potentials

    H. Beck and O. Marsalek, More converged, less accu- rate? Reassessing standard choices for ab initio water us- ing machine learning potentials, arXiv[physics.chem-ph], arXiv:2603.20451 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  33. [41]

    Schran, F

    C. Schran, F. L. Thiemann, P. Rowe, E. A. M¨ uller, O. Marsalek, and A. Michaelides, Machine learning poten- tials for complex aqueous systems made simple, Proc. Natl. Acad. Sci. U. S. A.118, e2110077118 (2021)

    2021

  34. [42]

    Singraber, T

    A. Singraber, T. Morawietz, J. Behler, and C. Dellago, Den- sity anomaly of water at negative pressures from first prin- ciples, J. Phys.: Condens. Matter30, 254005 (2018)

    2018

  35. [43]

    Cheng, J

    B. Cheng, J. Behler, and M. Ceriotti, Nuclear Quantum Effects in Water at the Triple Point: Using Theory as a Link Between Experiments, J. Phys. Chem. Lett.7, 2210 (2016)

    2016

  36. [44]

    Kapil, D

    V. Kapil, D. M. Wilkins, J. Lan, and M. Ceriotti, Inexpensive modeling of quantum dynamics using path integral gener- alized Langevin equation thermostats, J. Chem. Phys.152, 124104 (2020)

    2020

  37. [45]

    Omranpour, P

    A. Omranpour, P. Montero De Hijes, J. Behler, and C. Del- lago, Perspective: Atomistic simulations of water and aque- ous systems with machine learning potentials, J. Chem. Phys.160, 170901 (2024)

    2024

  38. [46]

    P. M. Piaggi, A. Z. Panagiotopoulos, P. G. Debenedetti, and R. Car, Phase Equilibrium of Water with Hexagonal and Cubic Ice Using the SCAN Functional, J. Chem. Theory Comput.17, 3065 (2021)

    2021

  39. [47]

    Zhang, Phase Diagram of a Deep Potential Water Model, Phys

    L. Zhang, Phase Diagram of a Deep Potential Water Model, Phys. Rev. Lett.126, 10.1103/PhysRevLett.126.236001 (2021)

    work page doi:10.1103/physrevlett.126.236001 2021

  40. [48]

    K. Xu, Y. Hao, T. Liang, P. Ying, J. Xu, J. Wu, and Z. Fan, Accurate prediction of heat conductivity of water by a neu- roevolution potential, J. Chem. Phys.158, 204114 (2023)

    2023

  41. [49]

    Riemelmoser, C

    S. Riemelmoser, C. Verdi, M. Kaltak, and G. Kresse, Ma- chine Learning Density Functionals from the Random-Phase Approximation, J. Chem. Theory Comput.19, 7287 (2023)

    2023

  42. [50]

    Hilpert and G

    T. Hilpert and G. Kresse, Accurate thermophysical prop- erties of water using machine-learned potentials, arXiv [physics.chem-ph], arXiv:2601.21103 (2026)

    work page arXiv 2026

  43. [51]

    Kuryla, F

    D. Kuryla, F. Berger, G. Cs´ anyi, and A. Michaelides, How accurate are DFT forces? Unexpectedly large uncertainties in molecular datasets, J. Chem. Phys.163, 224313 (2025)

    2025

  44. [52]

    O’Neill, B

    N. O’Neill, B. X. Shi, W. J. Baldwin, W. C. Witt, G. Cs´ anyi, J. D. Gale, A. Michaelides, and C. Schran, Towards rou- tine condensed phase simulations with delta-learned cou- pled cluster accuracy: Application to liquid water, J. Chem. Theory Comput.21, 11710 (2025)

    2025

  45. [54]

    S. J. Clark, M. D. Segall, C. J. Pickard, P. J. Hasnip, M. I. J. Probert, K. Refson, and M. C. Payne, First principles meth- ods using CASTEP, Z. Kristallogr. - Cryst. Mater.220, 567 (2005)

    2005

  46. [55]

    T. D. K¨ uhne, M. Iannuzzi, M. Del Ben, V. V. Rybkin, P. Seewald, F. Stein, T. Laino, R. Z. Khaliullin, O. Sch¨ utt, F. Schiffmann, D. Golze, J. Wilhelm, S. Chulkov, M. H. Bani-Hashemian, V. Weber, U. Borˇ stnik, M. Taillefu- mier, A. S. Jakobovits, A. Lazzaro, H. Pabst, T. M¨ uller, R. Schade, M. Guidon, S. Andermatt, N. Holmberg, G. K. Schenter, A. Hehn...

    2020

  47. [56]

    V. Blum, R. Gehrke, F. Hanke, P. Havu, V. Havu, X. Ren, K. Reuter, and M. Scheffler,Ab Initiomolecular simula- tions with numeric atom-centered orbitals, Comput. Phys. Commun.180, 2175 (2009)

    2009

  48. [57]

    Q. Sun, T. C. Berkelbach, N. S. Blunt, G. H. Booth, S. Guo, Z. Li, J. Liu, J. D. McClain, E. R. Sayfutyarova, S. Sharma, S. Wouters, and G. K.-L. Chan, PySCF: The Python-based simulations of chemistry framework, Wiley In- terdiscip. Rev.:Comput. Mol. Sci.8, e1340 (2018)

    2018

  49. [58]

    Giannozzi, S

    P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. D. Corso, S. de Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Maz- zarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbrac- cia, S. ...

    2009

  50. [60]

    Kresse and J

    G. Kresse and J. Furthm¨ uller, Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set, Comput. Mater. Sci.6, 15 (1996)

    1996

  51. [62]

    M. Holz, S. R. Heil, and A. Sacco, Temperature-dependent self-diffusion coefficients of water and six selected molecular liquids for calibration in accurate 1H NMR PFG measure- ments, Phys. Chem. Chem. Phys.2, 4740 (2000)

    2000

  52. [64]

    A. D. Becke and E. R. Johnson, A density-functional model of the dispersion interaction, J. Chem. Phys.123, 154101 (2005)

    2005

  53. [66]

    Q. Sun, X. Zhang, S. Banerjee, P. Bao, M. Barbry, N. S. Blunt, N. A. Bogdanov, G. H. Booth, J. Chen, Z.-H. Cui, J. J. Eriksen, Y. Gao, S. Guo, J. Hermann, M. R. Hermes, K. Koh, P. Koval, S. Lehtola, Z. Li, J. Liu, N. Mardirossian, J. D. McClain, M. Motta, B. Mussard, H. Q. Pham, A. Pulkin, W. Purwanto, P. J. Robinson, E. Ronca, E. R. Sayfutyarova, M. Sche...

    2020

  54. [68]

    How reproducible are first-principles simulations of liquid water?

    A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. Bjerre Jensen, J. Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaard, J. Bergmann Maronsson, T. Maxson, T. Olsen, L. Pastewka, A. Peterson, C. Rostgaa...

    2017

  55. [69]

    Machine learning interatomic potential 3 B

    Protocol for reliable simulations of liquids 3 A. Machine learning interatomic potential 3 B. Delta-learning MLIP 5 C. Molecular dynamics computational details 6

  56. [70]

    Validation againstab initioMD 8

  57. [71]

    FHI-aims 10 B

    Electronic structure details 9 A. FHI-aims 10 B. VASP 10 C. Quantum Espresso 10 D. CP2K 10 E. PySCF 11 F. CASTEP 11 G. D3 11

  58. [72]

    Nuclear quantum effects 12

    Final density and diffusion estimates 12 A. Nuclear quantum effects 12

  59. [73]

    Literature compilation 15

  60. [74]

    Effect of D3 dispersion variants on density and diffusion 17

  61. [75]

    Basis set incompleteness of the TZV2P-GTH basis set 18 B

    Basis set and pseudopotential errors 18 A. Basis set incompleteness of the TZV2P-GTH basis set 18 B. Inconsistency errors in the GTH pseudopotentials 19

  62. [76]

    Radial distribution function 21 B

    Validation tests 20 A. Radial distribution function 21 B. WATER27 interaction energies 22 References 24 ∗ oneilln@mpip-mainz.mpg.de; These authors contributed equally to this work. † mail@benjaminshi.com; These authors contributed equally to this work. 2

  63. [77]

    We will start by discussing the architecture and dataset of our machine-learning interatomic potential (MLIP), which achieves state-of-the-art performance

    PROTOCOL FOR RELIABLE SIMULATIONS OF LIQUIDS We will describe and summarize the protocol we have developed to ensure reliable and repro- ducible simulations of liquids here. We will start by discussing the architecture and dataset of our machine-learning interatomic potential (MLIP), which achieves state-of-the-art performance. Then, we will discuss furth...

  64. [78]

    kBT 6ı”L (1) where

    Python package, with a timestep of 0.5 fs at 298 K. A box size of 64 waters was used to be consistent with most previous literature. Pressure was controlled in NPT simulations using a Nos´ e-Hoover-type barostat, with an external isotropic pressure of 1 bar applied using a barostat relaxation time of 500 fs. The mean density was computed from the block av...

  65. [79]

    The most rigorous test is to compare toab initiosimulations

    VALIDATION AGAINSTAB INITIOMD While MLIPs can largely alleviate the sampling bottleneck highlighted in the Introduction of the main text, to obtain reliable thermodynamic observables, the question remains whether the underlying model can itself reproduce the ground truth. The most rigorous test is to compare toab initiosimulations. We performedab initiomo...

  66. [80]

    ELECTRONIC STRUCTURE DETAILS We discuss the code-specific electronic structure parameters in this section. They are sum- marized in Table 6, where we provide settings for both ‘standard’ settings (either commonly used in the literature or the defaults recommended by developers) and the tight ‘converged’ settings we have used to reach the high precision ne...

  67. [81]

    We tabulate these final estimates for the density and diffusion coefficient in Tables 7 and 8, respectively

    FINAL DENSITY AND DIFFUSION ESTIMATES The left and middle panels of Figure 1 of the main text graphically illustrates the final predicted diffusion and density predictions for the ‘converged’ and ‘standard’ numerical settings, respec- tively. We tabulate these final estimates for the density and diffusion coefficient in Tables 7 and 8, respectively. TABLE...

  68. [82]

    Tables 10 and 11 below summarise the technical details of each work, including the code used, method (ab initioor MLIP based), plane-wave cutoff and basis set size (where relevant)

    LITERATURE COMPILATION The right panel of Figure 1 of the main text graphically illustrates the range in predicted diffusion and density predictions over the years, by various different codes and approaches. Tables 10 and 11 below summarise the technical details of each work, including the code used, method (ab initioor MLIP based), plane-wave cutoff and ...

    2017

  69. [83]

    TABLE 12:Effect of the Axilrod–Teller–Muto (ATM) [47] term on the computed density and self- diffusion coefficient for Becke-Johnson and zero-damping schemes

    EFFECT OF D3 DISPERSION VARIANTS ON DENSITY AND DIFFUSION While the standard choice of D3 dispersion is zero-damping, with 2-body terms, in Table 12 we also compare the density and diffusion coefficient of the other major damping scheme – Becke- Johnson, along with the Axilrod-Teller-Muto term to capture 3-body terms in the dispersion. TABLE 12:Effect of ...

  70. [84]

    BASIS SET AND PSEUDOPOTENTIAL ERRORS As discussed in the main text, and shown in Tables 10 and 11 in Section 5, the majority of the previous literature employ the TZV2P basis set and GTH-PBE pseudopotential. This raises questions on the origins of their inaccuracies and we will show in this section it comes from both basis set incompleteness of the TZV2P ...

  71. [85]

    First, we will compare radial distribution functions (with classical nuclei) between the different 20 codes — another important hallmark of liquid water structural parameters

    VALIDATION TESTS In this section, we highlight some additional validation tests performed within this study. First, we will compare radial distribution functions (with classical nuclei) between the different 20 codes — another important hallmark of liquid water structural parameters. Subsequently, we will compare the WATER27 dataset of water cluster bindi...

  72. [86]

    Meanwhile the RDFs with standard numerical settings show some wider variation (which is well correlated with the extent of the discrepancies in the diffusion coefficent.) FIG

    and neutron scattering [50]. Meanwhile the RDFs with standard numerical settings show some wider variation (which is well correlated with the extent of the discrepancies in the diffusion coefficent.) FIG. 5:Comparison of O-O, O-H and H-H RDFs predicted by the standard and converged settings of each code. 21 B. WATER27 interaction energies In Table 16, we ...

  73. [87]

    O’Neill, B

    N. O’Neill, B. X. Shi, W. J. Baldwin, W. C. Witt, G. Cs´ anyi, J. D. Gale, A. Michaelides, and C. Schran, Towards routine condensed phase simulations with delta-learned coupled cluster accu- racy: Application to liquid water, J. Chem. Theory Comput.21, 11710 (2025)

    2025

  74. [88]

    Montero de Hijes, C

    P. Montero de Hijes, C. Dellago, R. Jinnouchi, and G. Kresse, Density isobar of water and melting temperature of ice: Assessing common density functionals, J. Chem. Phys.161, 131102 (2024)

    2024

  75. [89]

    Batatia, D

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csanyi, MACE: Higher order equivariant message passing neural networks for fast and accurate force fields, Advances in Neural Information Processing Systems35, 11423 (2022)

    2022

  76. [90]

    I. J. Parker, M. J. Hoffmann, W. J. Baldwin, S. Han, S. Gupta, K. D. Fong, A. Michaelides, C. Schran, S. De, and G. Cs´ anyi, False metallization in short-ranged machine learned interatomic potentials, arXiv[physics.chem-ph], arXiv:2603.04228 (2026)

    work page arXiv 2026

  77. [91]

    W. J. Baldwin, I. Batatia, M. Vondr´ ak, J. T. Margraf, and G. Cs´ anyi, Design Space of Self– Consistent Electrostatic Machine Learning Interatomic Potentials, arXiv[physics.chem-ph], arXiv:2603.14700 (2026)

    work page arXiv 2026

  78. [92]

    Hjorth Larsen, J

    A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. Bjerre Jensen, J. Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaard, J. Bergmann Maronsson, T. Maxson, T. Olsen, L. Pastewka, A. Peterson, C. Rostgaa...

    2017

  79. [93]

    Bussi, D

    G. Bussi, D. Donadio, and M. Parrinello, Canonical sampling through velocity rescaling, J. Chem. Phys.126, 014101 (2007)

    2007

  80. [94]

    Yeh and G

    I.-C. Yeh and G. Hummer, System-size dependence of diffusion coefficients and viscosities from molecular dynamics simulations with periodic boundary conditions, J. Phys. Chem. B108, 15873 (2004)

    2004

Showing first 80 references.