Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

A many-body machine-learned force field trained on revPBE-D3 data reproduces the experimental ion-specific anomaly of water diffusion and shows that Na+ retards while I− accelerates.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 20:39 UTC pith:HCBNR4AY

load-bearing objection Solid MACE-revPBE-D3 work that improves the NaCl Dw/D0 curve over DeePMD and supplies a clean pure-shell/overlap decomposition; the functional limitation is already flagged and does not break the internal claim. the 2 major comments →

arxiv 2604.13659 v3 pith:HCBNR4AY submitted 2026-04-15 physics.chem-ph cond-mat.soft

Ion-Specific Anomalous Water Diffusion in Aqueous Electrolytes: A Machine-Learned Many-Body Force Field Study with MACE

Massimo Ciacchi , Ilnur Saitov , Nico Di Fonte , Isabella Daidone , Carlo Pierleoni This is my paper
classification physics.chem-ph cond-mat.soft
keywords anomalous water diffusionaqueous electrolytesmachine-learned force fieldMACErevPBE-D3hydration shellsion-specific effectspotentials of mean force
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Water molecules move more slowly when NaCl is dissolved and more quickly when CsI is dissolved—an ion-specific anomaly that classical force fields have long failed to capture. This paper trains an equivariant many-body machine-learned force field (MACE) on density-functional theory data (revPBE-D3) and uses it for nanosecond-scale classical molecular dynamics of both salts over a wide concentration range. The simulations recover the experimental trends and improve quantitatively on earlier machine-learned models trained on the same electronic-structure method, especially for the NaCl slowdown. The improvement is traced to a deeper free-energy well for Na+–water and a measurable retarding effect that extends into Na+’s second hydration shell; the CsI speedup is shown to be driven mainly by the diffuse, weakly bound shell of I−. Shell-resolved diffusivities and ion–oxygen potentials of mean force supply a coherent microscopic picture of how structure-making and structure-breaking ions reshape water mobility.

Core claim

Classical molecular dynamics driven by a MACE force field trained on revPBE-D3 energies, forces and stresses quantitatively reproduces the experimentally observed concentration dependence of the relative water diffusion coefficient—suppression in NaCl, enhancement in CsI—and improves on DeePMD results obtained with the same functional, the gain arising from a stronger Na+–water interaction in the first shell plus a non-negligible retarding contribution of the second hydration shell of Na+, while the CsI acceleration is primarily driven by the diffuse hydration shell of I−.

What carries the argument

Shell-decomposition of short-time water diffusivities (pure first shells, overlapping shells, and second shells) together with the corresponding ion–oxygen potentials of mean force; these quantities map local free-energy barriers onto the measured mobility changes.

Load-bearing premise

The claim rests on the premise that the revPBE-D3 functional, known to underestimate the height of the Na–O first peak relative to higher-level electronic-structure methods, is still accurate enough that residual errors do not reverse the sign or ranking of the ion-specific diffusion anomalies.

What would settle it

Recompute the same concentration series of relative water diffusion coefficients with a MACE (or equivalent) force field trained on RPA or MP2 data for the same ions; if the NaCl slowdown or CsI speedup disappears or reverses relative to experiment, the central claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The manuscript trains a MACE equivariant ML force field on revPBE-D3 energies, forces and stresses for pure water and aqueous NaCl/CsI, then uses classical MD (size-extrapolated Green–Kubo) to compute water self-diffusion and shear viscosity over 0.89–3.56 mol/kg. It reproduces the experimental ion-specific anomaly (Dw/D0 < 1 for NaCl, > 1 for CsI) and reports quantitative improvement over DeePMD trained on the same functional, especially for NaCl. The improvement is attributed to a deeper Na+–O free-energy well and a measurable retarding contribution from Na+’s second hydration shell; for CsI the acceleration is assigned primarily to the diffuse I− shell. Supporting evidence includes RDFs, neutron structure factors, hydrogen-bond counts, shell-decomposed short-time diffusivities, and ion–oxygen PMFs.

Significance. If the results hold, the work supplies a concrete, architecture-level demonstration that higher-order equivariant message-passing (MACE) can improve ion–water free-energy landscapes relative to DeePMD at fixed DFT theory, and that this improvement is dynamically consequential for the long-standing anomalous-diffusion problem. The size-extrapolated transport coefficients, block-error analysis, and direct side-by-side comparison with experiment and with DeePMD on the same functional constitute a reproducible benchmark that future MLFFs (including those trained beyond GGA) can be measured against. The shell-decomposition and PMF analysis also give a clear microscopic picture that classical non-polarizable models have systematically missed.

major comments (2)
  1. The central claim of quantitative improvement over DeePMD rests on a deeper Na+–O PMF (Fig. 17, Tables S6–S7) and second-shell retardation (Fig. 8d, Table S12). The manuscript itself cites O’Neill et al. that revPBE-D3 underestimates the Na–O first-peak height relative to RPA/MP2. Because the training data and the DeePMD reference share this functional, the residual error could still reverse the NaCl ranking once higher-level data are used. A short, explicit discussion of this risk (or a limited higher-level single-point check on representative Na–O configurations) is needed to bound the claim.
  2. AIMD validation trajectories are only ~10 ps (Appendix B). While structural RDFs and short-time VACFs are compared (Figs. 10–13), the transport coefficients that form the paper’s main result require nanosecond sampling. The authors should state more clearly that the AIMD comparison validates only local structure and short-time dynamics, not the long-time Dw/D0 values themselves.
minor comments (4)
  1. Fig. 3 ion–ion RDFs at low concentration are noisy; a brief note that the noise is statistical (few ions) would help readers.
  2. Hydration-shell cut-offs are defined via the inflection of n(r) (Fig. S3). The numerical values should also appear in the main text or a table for reproducibility.
  3. The multi-head fine-tuning protocol (30 epochs, 10 k foundation configurations) is described only in the Appendix; a one-sentence summary in §II would improve accessibility.
  4. Typographical inconsistencies appear in a few places (e.g., “Nos´ e-Hoover”, “˚A”); a light copy-edit pass is warranted.

Circularity Check

0 steps flagged

No significant circularity: DFT-trained MACE MD yields independent transport predictions benchmarked on external experiment and prior DeePMD.

full rationale

The force field is trained exclusively on revPBE-D3 energies, forces and stresses (Appendix A/C); water diffusion and viscosity are then obtained from Green–Kubo integrals of independent NVT trajectories (Eqs. 6–7) that were never used in training or fine-tuning. Experimental Dw/D0 and η/η0 (Müller & Hertz, Jones & Fornwalt, etc.) serve only as external benchmarks (Figs. 6–7, Tables S3–S5). The claimed improvement over DeePMD (same functional) is a post-hoc structural comparison of Na–O PMFs and shell-resolved diffusivities (Figs. 8–9, 17; Tables S6–S7, S12), not a quantity fitted to the target anomaly. Self-citations to Avula et al. and Ding et al. supply context and baselines but are not load-bearing uniqueness claims; the derivation chain remains open and falsifiable against higher-level theory or experiment. No self-definitional identities, fitted-input-as-prediction steps, or ansatz-smuggling appear.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 0 invented entities

The central claim rests on the adequacy of a single GGA functional, on classical nuclear dynamics, on a particular neural-network architecture and training protocol, and on operational definitions of hydration shells. No new physical entities are postulated; free parameters are the usual ML and analysis choices.

free parameters (3)
  • MACE architecture hyper-parameters (128 channels, L=1, correlation order 3, lmax=3, 6 Å cutoff, energy/force/stress weig
    Chosen by the authors; different choices would alter force accuracy and therefore the predicted diffusivities.
  • Hydration-shell cut-offs (Na 3.2 Å, Cl 3.8 Å, Cs 4.0 Å, I 4.3 Å) defined via inflection of n(r)
    Operational definition used for all shell-decomposed diffusivities; small shifts change the pure-shell populations and the reported relative Dw values.
  • Fine-tuning protocol (30 epochs multi-head replay with 10 k foundation-model configurations)
    Required to eliminate unphysical Na–Na and Cs–Cs pairing present in the scratch-trained M1 model; the final FT-M1 results depend on this ad-hoc correction.
axioms (3)
  • domain assumption revPBE-D3 DFT energies, forces and stresses constitute a sufficiently accurate ground truth for ion–water interactions that the sign of the diffusion anomaly is preserved.
    Stated in Abstract and §I; known to underestimate Na–O peak height (O’Neill et al. cited by the authors).
  • domain assumption Classical nuclear dynamics on the ML potential surface is adequate; nuclear quantum effects may be neglected for the relative diffusivities.
    Implicit throughout; never tested.
  • standard math Green–Kubo integrals of VACF and stress autocorrelation, after finite-size extrapolation, equal the macroscopic transport coefficients.
    Standard statistical-mechanics identity used in §III B.

pith-pipeline@v1.1.0-grok45 · 28628 in / 2741 out tokens · 30364 ms · 2026-07-12T20:39:19.128337+00:00 · methodology

0 comments
read the original abstract

The dynamics of water in electrolyte solutions exhibits a striking, ion-specific anomaly: the diffusion coefficient of water is enhanced relative to the neat liquid in chaotropic CsI solutions, yet suppressed in kosmotropic NaCl solutions. This phenomenon, long challenging for classical force-field-based molecular dynamics, is studied here using classical molecular dynamics simulations with a many-body machine-learned force field (MLFF) trained within the MACE equivariant graph neural network framework. The force field is trained on energies, forces, and stresses computed at the density functional theory level with the revPBE-D3 exchange--correlation functional, which provides a reliable balance between accuracy and computational efficiency for aqueous systems. Simulations of NaCl and CsI aqueous solutions at ambient conditions over a concentration range of 0.89--3.56 mol/kg reproduce the experimentally observed anomalous diffusion and show a quantitative improvement over previous results obtained with the DeePMD framework, trained on the same theory, particularly for NaCl solutions. This improvement is traced to a stronger Na$^{+}$--water interaction in the first hydration shell and the non-negligible retarding contribution of the second hydration shell of Na$^{+}$. For CsI solutions, the water acceleration is shown to be primarily driven by the anion I$^{-}$, whose diffuse and weakly structured hydration shell facilitates rapid water exchange with the bulk. These results are rationalised through a shell-decomposition analysis of time-dependent water diffusivities and ion--oxygen potentials of mean force providing a coherent microscopic picture of the acceleration--retardation mechanism in the studied aqueous electrolytes.

Figures

Figures reproduced from arXiv: 2604.13659 by Carlo Pierleoni, Ilnur Saitov, Isabella Daidone, Massimo Ciacchi, Nico Di Fonte.

Figure 1
Figure 1. Figure 1: FIG. 1: Density as a function of concentration (EoS) at [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Radial distribution functions (RDFs) involving the oxygen as a function of concentration for NaCl and CsI [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Ion-ion radial distribution functions (RDFs) as a function of concentration. Systems with 250 water [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Total neutron scattering structure factor [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Reduced structure factor for NaCl (upper [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Relative diffusion dependence on the [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Panel (a): Values of relative diffusion at 2.5 ps for total first solvation shells (pure first solvation shells + [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Potential of Mean Force: concentration [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Oxygen RDF comparison between DFT [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: RDF comparison between DFT, M1 and [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13: (a) Velocity Autocorrelation Function, (b) [PITH_FULL_IMAGE:figures/full_fig_p011_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14: Comparison between DFT and FT-M1 Model [PITH_FULL_IMAGE:figures/full_fig_p011_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15: Oxygen-Oxygen RDFs comparison with Avula [PITH_FULL_IMAGE:figures/full_fig_p012_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16: Radial distribution function comparison [PITH_FULL_IMAGE:figures/full_fig_p012_16.png] view at source ↗
Figure 19
Figure 19. Figure 19: FIG. 19: The dependence of the diffusion coefficient of [PITH_FULL_IMAGE:figures/full_fig_p013_19.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18: The dependence of the diffusion coefficient of [PITH_FULL_IMAGE:figures/full_fig_p013_18.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. VASP Plugins: Linking the Vienna ab-initio Simulation Package with Python

    cond-mat.mtrl-sci 2026-07 accept novelty 5.5

    A C++/pybind11 shared-memory plugin layer exposes VASP SCF and ionic data as NumPy arrays so Python can modify structure, forces, local potential, and occupancies in place.

Reference graph

Works this paper leans on

3 extracted references · cited by 1 Pith paper

  1. [1]

    Dissolving salt is not equivalent to applying a pressure on water

    Zhang, Chunyi and Yue, Shuwen and Panagiotopoulos, Athanassios Z. and Klein, Michael L. and Wu, Xifan, “Dissolving salt is not equivalent to applying a pressure on water”, Nat. Commun.��1, 822 (2022)

    2022

  2. [2]

    Importance of van der Waals effects on the hydration of metal ions from the Hofmeister series

    Zhou, Liying and Xu, Jianhang and Xu, Limei and Wu, Xifan, “Importance of van der Waals effects on the hydration of metal ions from the Hofmeister series”, J. Chem. Phys.���12, 124505 (2019)

    2019

  3. [3]

    Understanding the Anomalous Diffusion of Water in Aqueous Electrolytes Using Machine Learned Potentials

    Avula, Nikhil V. S. and Klein, Michael L. and Balasubramanian, Sundaram, “Understanding the Anomalous Diffusion of Water in Aqueous Electrolytes Using Machine Learned Potentials”, J. Phys. Chem. Lett.��, 9500-9507 (2023)

    2023