REVIEW 3 major objections 4 minor 29 references
How important is the dielectric constant in water modeling? Evaluation of the performance of the TIP4P/$\varepsilon$ force field and its compatibility with the Joung-Cheatham NaCl model
T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read A water model tuned to the dielectric constant matches TIP4P/2005 across most properties and, without any charge rescaling, reproduces NaCl solution densities.
desk verdict Solid pure-water benchmark showing TIP4P/ε ties TIP4P/2005 while reproducing ε; the electrolyte causality claim is softer than the abstract suggests. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is the Kirkwood-factor decomposition of the static dielectric constant, $\varepsilon = 1 + \frac{4\pi\rho}{3 k_B T}\mu^2 g_K$, which splits ε into a single-molecule dipole μ and a collective orientational correlation factor $g_K$. TIP4P/ε achieves its dielectric accuracy by placing the negative dummy site much closer to the oxygen than TIP4P/2005 does ($d_{\mathrm{OM}} = 0.105$ Å versus 0.1546 Å), which raises the molecular dipole only about 5% (2.43 D versus 2.31 D) yet changes liquid-state orientational correlations enough to raise $g_K$ from 3.2 to 3.85 at ambient conditions. The paper pairs this identity with the standard benchmark used to rate rigid water models and with the unmodified Joung–Cheatham ion model to test transferability. The same identity also exposes the model's weak spot: in ice Ih at 273 K the Kirkwood factor collapses to 2.0, so TIP4P/ε still predicts the wrong trend of ε on freezing, like other point-charge models.
What would settle it
Re-fit the Na⁺–O and Cl⁻–O cross parameters self-consistently for TIP4P/ε from crystal and melt data and rerun the molality–density curves; if the densities shift by more than about 1%, the reported agreement was inherited from the original cross parameters. Alternatively, measure the ambient Kirkwood factor $g_K$ experimentally: if the true value is near 3.2 rather than 3.85, TIP4P/ε achieves its dielectric constant through incorrect orientational correlations.
Extended reading notes
Core claim
The paper's central claim is that TIP4P/ε, a rigid four-point water model parametrized to reproduce the experimental dielectric constant, performs as well as TIP4P/2005 for most pure-water and ice properties and, because of its accurate dielectric response, transfers to electrolyte solutions without further fitting. Concretely, the two models earn nearly equal overall scores on the standard benchmark (7.59 for TIP4P/2005 versus 7.54 for TIP4P/ε, with OPC at 6.26); TIP4P/ε reproduces the static dielectric constant of liquid water over a wide temperature range at 1 bar, at saturation, and at 500 bar; and unmodified Joung–Cheatham NaCl at full integer charges yields solution densities and temperatures of maximum density close to experiment. The dielectric accuracy comes not from large charges (TIP4P/ε charges are only about 5% smaller than TIP4P/2005's) but from stronger orientational correlations, reflected in a Kirkwood factor $g_K \approx 3.85$ versus 3.2. The authors conclude that a robust parametrization can be achieved without charge scaling, while acknowledging that the same point-charge limitations reappear in solution transport, with JC–TIP4P/ε viscosities far too large.
Load-bearing premise
The electrolyte conclusion rests on the assumption that the Joung–Cheatham ion–water cross-interaction parameters used with TIP4P/ε remain valid for that model even though they were fitted for a different host water model; if those cross terms already compensate for the dielectric deficiency of their original host, the agreement with experiment would not isolate the dielectric constant as the cause.
Editorial extensions
If this is right
- TIP4P/ε can be used as a rigid water model where accurate static dielectric response matters, matching TIP4P/2005 on densities, transport, ice polymorph densities, and surface tension while fixing the roughly 25% dielectric deficit.
- Joung–Cheatham NaCl at full integer charges, with no charge rescaling and no Lorentz–Berthelot adjustment, gives acceptable solution densities and temperatures of maximum density in TIP4P/ε, at a level comparable to a specifically parametrized scaled-charge electrolyte force field.
- For phase-boundary studies, TIP4P/2005 remains the better choice, since it predicts melting and critical points more accurately; TIP4P/ε melts at about 238 K.
- Electrolyte transport remains a shared failure of both point-charge models: JC–TIP4P/ε viscosities are far too high, pointing to missing local charge fluctuations as the next bottleneck.
- The benchmark scores imply that OPC, despite its good dielectric constant, is clearly worse overall, so dielectric accuracy alone does not guarantee a good water model.
Reading between the lines
- A testable extension of the paper's logic: if accurate ε is the causal factor, other rigid water models with accurate dielectric constants should also inherit the unmodified-JC-ion transferability; repeating the NaCl-density calculation with OPC water would probe this directly.
- A conservative caveat: the JC ion–water cross parameters used here were fitted for a different host water model, and the paper does not state which variant; if that host had a low dielectric constant, the cross terms may already encode compensation, so the observed agreement would not isolate ε as the cause.
- The Kirkwood-factor difference suggests a falsifiable physical picture: TIP4P/ε implies larger orientational correlations in ambient water than TIP4P/2005, and any experiment able to constrain $g_K$ could discriminate which model has the right collective structure.
- A practical extension would be to map the dielectric decrement of JC–TIP4P/ε over the full molality range, since the reported solution densities are good at low concentration while the concentrated regime is exactly where the viscosity failure appears.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This manuscript evaluates the TIP4P/epsilon water model against TIP4P/2005 and OPC using the Vega-Abascal benchmark for pure-water properties, and then tests its transferability to NaCl electrolyte solutions using the Joung-Cheatham (JC) ion force field. The authors report that TIP4P/epsilon matches TIP4P/2005 overall accuracy on the benchmark, largely outperforms OPC, and, when combined with full-charge JC ions, gives good solution densities and dielectric constants without charge rescaling or reparameterization, which they interpret as evidence that accurate dielectric constants are a valuable target property in water modeling.
Significance. If the results hold, they challenge a common interpretation of the charge-scaling paradigm by showing that a rigid point-charge water model can simultaneously reproduce the dielectric constant and remain competitive with TIP4P/2005 on a wide range of properties, while also providing reasonable electrolyte thermodynamics with unmodified full-charge ions. The paper includes a substantial amount of original simulation data, particularly for dielectric constants, transport properties, melting point, ice polymorph densities, and electrolyte density/TMD, and the direct-coexistence melting point calculation for TIP4P/epsilon is a useful addition. The comparison with OPC and TIP4P/2005 is framed within a well-established benchmark, and the authors are candid about limitations, including the failure of solution viscosities.
major comments (3)
- [V.A, Table IV] The central causal claim--that reproducing the dielectric constant is what allows full-charge JC ions to give acceptable NaCl solution densities--is not isolated by the presented simulations because Table IV does not state which water-model variant of the JC force field was used for the Na+-O and Cl--O cross interactions. JC parameters are host-water specific, and the manuscript's characterization of the ions as 'targeted to their crystal and melt properties only' is incomplete if the cross terms originate from a different JC water parameter set. Since Fig. 7(a) compares JC-TIP4P/epsilon against JC-TIP4P/2005 using different water-ion cross interactions (Table IV vs. Table II of Ref. 89), the observed density improvement could arise from cross-term provenance rather than from the dielectric constant of the host water. The authors must specify the exact JC water-model variant used, justify its choice, and ideally run a control where the cross interactions are kept fixed while only the water model changes, or at least discuss why such a control is not conclusive.
- [IV.F, Table III] The conclusion that TIP4P/epsilon and TIP4P/2005 'perform equally well' rests on a comparison in which the TIP4P/2005 and OPC values are taken from Ref. 41 rather than recomputed with the same simulation protocols, and several TIP4P/epsilon values are taken from the original parametrization paper (marked with asterisks). This makes the reported benchmark scores (7.59 vs 7.54) appear more precise than the underlying data support. The authors should state the estimated statistical uncertainties in the VA scores and clarify whether the literature values were obtained with the same treatment of long-range corrections, system sizes, and ensemble conditions; otherwise the claim of equal performance is only as strong as the consistency of the compiled references.
- [V.E, Conclusions] The paper reports that JC-TIP4P/epsilon yields viscosities that are far too large across the entire electrolyte concentration range (Fig. 9), yet the conclusions still refer to the model as 'a very robust model for preliminary studies of solution properties.' This is not internally contradictory, but the transport failure is a significant limitation that should be weighed more explicitly in the summary, particularly because the abstract highlights the advantage of dielectric constants for solution properties without mentioning that the dynamic properties are not transferable. The authors should temper the 'robust model' claim or explicitly state the domain of applicability (thermodynamic properties only) in the abstract and conclusions.
minor comments (4)
- [II, Eq. (6)] The non-local screening model in Eqs. (4)-(6) is presented as a physical motivation for charge scaling, but it is not used in the simulations and is admittedly a 'mere caricature.' This section could be tightened to avoid giving the impression that the electrolyte results depend on this model; consider moving it to a more clearly labeled conceptual discussion or shortening it.
- [IV.A, Fig. 1] The comparison of dielectric constants for TIP4P/2005 and TIP4P/epsilon would benefit from error bars or shaded uncertainty bands on the simulation data, especially since the models are claimed to differ by about 25% and the reader cannot assess sampling uncertainty from the figure.
- [III.D, Eq. (10)] The Yeh-Hummer correction is applied to the diffusion coefficient, but the text does not state which viscosity value is used for the correction in each state point; please specify whether the simulated or experimental viscosity was used.
- [General] There are several typographical and formatting issues, including inconsistent rendering of 'TIP4P/epsilon' as 'TIP4P ε' in places, the use of 'gK' vs. 'g_K', and an apparent duplicate entry for Ref. 77; a careful proofread is needed.
Circularity Check
No significant circularity: external benchmarks and inherited parameters.
full rationale
The manuscript contains no step in which a claimed prediction reduces by construction to a fitted input or to a self-citation chain. TIP4P/epsilon parameters (Table II) are inherited from Ref. 39 and are not refitted in this work; the paper explicitly attributes them. The Vega-Abascal scores in Table III compare TIP4P/epsilon values computed by the authors (or taken from Ref. 39) against the external benchmark of Ref. 16, with TIP4P/2005 and OPC values taken from Ref. 41; none of the rated properties is fitted to the experimental targets being scored. The dielectric constant of TIP4P/epsilon is a target property of the original model, so reporting epsilon at further state points is a transferability check rather than a first-principles derivation; the paper does not claim to derive epsilon from the model. Electrolyte densities use Joung-Cheatham parameters (Ref. 30) that were not optimized for TIP4P/epsilon, making the comparison an external transferability test rather than a fitted quantity. Self-references such as Refs. 27, 67, and 71 are used to motivate expectations and to document methodology (e.g., the Monte Carlo code for ice polymorphs and the dielectric-constant technique for ice), not as load-bearing proof of the central conclusion. The genuine caveats are evidentiary rather than circular: Table IV does not state which JC water-model variant supplies the NaCl-water cross-interactions, and the authors explicitly concede that the better electrolyte density performance 'could be, nevertheless, accidental' (Sec. V.B). Likewise, the solution dielectric-constant comparison is qualified by the conducting-solution caveat (Sec. V.D). These are confounding-factor and provenance concerns, not cases where an equation equals its input or a fitted parameter is renamed as a prediction. Accordingly, the derivation chain is self-contained with respect to the paper's own claims.
Assumptions & free parameters
free parameters (3)
- TIP4P/epsilon charges =
q(H)=+0.527 e; q(M)=-1.054 e
- TIP4P/epsilon LJ and geometry parameters =
sigma=3.165 A; epsilon=0.7732 kJ/mol; d_OH=0.9572 A; d_OM=0.105 A; theta=104.52 deg
- Joung-Cheatham NaCl parameters =
Na+: sigma=2.160 A, eps=1.4752 kcal/mol, q=+1; Cl-: sigma=4.830 A, eps=0.0536 kcal/mol, q=-1; plus Na-O and Cl-O cross…
assumptions (4)
- domain assumption Eq. (7) gives the static dielectric constant from dipole-moment fluctuations in an insulating periodic system.
- domain assumption The Vega-Abascal benchmark tolerances and weighting are a fair measure of water-model quality.
- domain assumption Joung-Cheatham ion-water cross parameters optimized for another TIP4P-class water model are transferable to TIP4P/epsilon without rescaling.
- ad hoc to paper The non-local screening model in Eqs. (4)-(6), with a Lorentzian form for n_infinity^2(k), captures distance-dependent screening relevant to charge scaling.
Cite this review
Pith. "Pith review of How important is the dielectric constant in water modeling? Evaluation of the performance of the TIP4P/$\varepsilon$ force field and its compatibility with the Joung-Cheatham NaCl model." pith.science (2026). https://pith.science/paper/MYMANEKN
@misc{pith2026250600153,
author = {Pith},
title = {Pith review of: How important is the dielectric constant in water modeling? Evaluation of the performance of the TIP4P/$\varepsilon$ force field and its compatibility with the Joung-Cheatham NaCl model},
year = {2026},
howpublished = {\url{https://pith.science/paper/MYMANEKN}},
note = {Machine review of arXiv:2506.00153}
}
abstract
Efficient large-scale computer simulations of aqueous solutions require the use of accurate but simple empirical force fields for water. However, the complexity of these systems evidences the difficulties in describing solution properties without due account of polarization. Different strategies to remedy this problem are parametrizing water force fields to the dielectric constant or charge scaling of solvated ions. In this work, we compare results from TIP4P/$\varepsilon$ and OPC models, which are parametrized to predict the dielectric constant, with results from TIP4P/2005, which is closer in spirit to the charge scaling strategy. The performance of the models is rated according to the Vega-Abascal benchmark. Our results show that TIP4P/$\varepsilon$ and TIP4P/2005 perform equally well, with the OPC model lying significantly behind. TIP4P/$\varepsilon$ can predict bulk phase properties (transport properties, thermal expansion coefficients, densities) of both liquid water and ice polymorphs, but also surface tensions, with an accuracy very similar to TIP4P/2005, while performing very well for dielectric constants over a wide range of pressures and temperatures. On the other hand, TIP4P/2005 provides a better description of phase boundaries, including liquid-vapor and freezing transitions. However, the accurate prediction of dielectric constants allows TIP4P/$\varepsilon$ to describe densities of NaCl solutions for models parametrized to their crystal and melt properties only. This is achieved without the need to rescale charges, modify the Lorentz-Berthelot rule or tune the ion's Lennard-Jones parameters. Our findings hinge on the significance of dielectric constants as a target property and show that a robust parametrization can be achieved without invoking the concept of charge scaling.
Figures
Figures from the paper (5 more)
Reference graph
Works this paper leans on
-
[1]
nsions, with an accuracy very similar to TIP4P/2005, while performing very well for dielectric constants over a wide range of pressures and temperatures. On the other hand, TIP4P/2005 provides a better description of phase boundaries, including liquid-vapor and freezing transitions. However, the accurate prediction of dielectric constants allows TIP4P/ε t...
work page 2005
-
[2]
Results are compared with the experimental data as reported in Ref.55
a) Infrared spectra of the OD stretch of an HDO molecule dissolved in H2O as predicted by TIP4P/2005 (blue) and TIP4P/ ε models. Results are compared with the experimental data as reported in Ref.55. b) Dependence of the OD stretch frequency with tempera- ture, compared with experimental results as compiled in Ref.78. The straight lines are linear fits ma...
work page 2005
-
[3]
It is clear that both TIP4P/2005 and TIP4P ε water models are quantitatively re- producing experimental results in a wide range of thermody- namic conditions. We therefore corroborate the observation of Azcatl and Alejandre that both the dielectric constant and TMD can be accurately reproduced by the same model. The question then remains to what extent ar...
work page 2005
-
[4]
b) Slopes of the melting/freezing modes plotted as a function of temperature
a) The evolution of number of liquid molecules in time for different temperatures around the melting point. b) Slopes of the melting/freezing modes plotted as a function of temperature. ure 4-b. The intersection of these rates with the x-axis pro- vides an estimated melting temperature of Tm,ε = 237.9 K. This value is smaller than the recently estimated m...
work page 2005
-
[5]
The data for the latter is taken from Ref
a) Relation of self-diffusion coefficient with the tem- perature at the atmospheric pressure p = 1 bar for the TIP4P/ ε and TIP4P/2005 water models. The data for the latter is taken from Ref
work page 2005
-
[6]
The data for the latter are taken from Ref
a) Relation of the shear viscosity with the temperature at the atmospheric pressure p = 1 bar for the TIP4P/ε and TIP4P/2005 water models. The data for the latter are taken from Ref
work page 2005
-
[7]
Solid line is the experimental data, taken from Ref
a) Density as a function of molality for the JC and Madrid-2019 NaCl force fields used in conjunction with TIP4P/2005 and TIP4P /ε water models. Solid line is the experimental data, taken from Ref
work page 2019
-
[8]
3.165 0.7732 0.9572 0.105 104.52 2.4345 2.174 H 0.527 M -1.054 TIP4P/2005 O
work page 2005
Show all 29 references
-
[9]
Experimental data are taken from Ref.75,76
Static dielectric constant as a function of temperature at p = 1 bar (a), p = psat (b), and p = 500 bar (c). Experimental data are taken from Ref.75,76. paper, Fuentes-Azcatl et al.39 argued that Lennard-Jones and Coulombic interactions behave independently. This allows us to ...
2005
-
[10]
effective
Here, two positive partial charges sitting on hydrogen positions are neutralized by a negative charge located at the center of a simple Lennard- Jones site, which accounts for the repulsion and dispersion of the oxygen atom. Alternatively, the negative charge can be shifted al...
1933 arXiv
-
[11]
It is evident that the TIP4P/ ε model re- produces experimental data very accurately, contrary to the TIP4P/2005 model, which predicts dielectric constants that deviate from the experimental values by more than 25 %. At first thought, one could suspect that such a large increa...
2005
-
[12]
From panel a) we see that TIP4P/ ε does a much better job than TIP4P/2005 at describing the low frequency side of the OD band, but has a significantly larger band width, and does therefore not reproduce the high frequency side of the band accurately. On the other hand, the spe...
2005
-
[16]
The model is then tested for transferability by studying so- 3 lution properties as described by the Joung-Cheatham elec- trolyte force field30, which was not specifically parametrized for TIP4P/ε. Our results show that TIP4P/ ε performs essen- tially as well as TIP4P/2005 for...
2005
-
[19]
c) Relation of the shear viscosity model with pressure for the TIP4P/ε at three different temperatures
b) mag- nified region of part a). c) Relation of the shear viscosity model with pressure for the TIP4P/ε at three different temperatures. Experimen- tal results are taken from Ref.86,87 . F. Comparison of the models So far, for selected properties, we have seen that the TIP4P/...
2005
-
[23]
Figure 8-a
Although this contribution appears to be relatively small,97 a direct com- parison with experimental data shown should be taken with some caution. Figure 8-a. shows that TIP4P/ε can predict very accurately a sharp drop of the dielectric constant with salt concentration, provid...
2005
-
[25]
Blazquez et al
This explains why the Madrid-2019 force field, with ionic charges scaled to 0.85 times the elemen- tary charge performs significantly better. Blazquez et al. 25 have shown that the discrepancy can be removed by further scaling the charges to even smaller values (equal to ±0.75...
2019
-
[26]
Indeed, these two surfaces could be different but are definitively not independent
This concept can be useful to bring predicted dielectric constants into agreement with experiment 23,27, but must be exercised with great cau- tion. Indeed, these two surfaces could be different but are definitively not independent. Linear response theory implies that a molecu...
2005
-
[27]
The solid line shows experimental results Property Exp
Viscosity as a function of molality for NaCl solution. The solid line shows experimental results Property Exp. JC-TIP4P/ ε Madrid-2019 ρmelt (kg/m3) at T = 1073.8 K 1556 1410 1331 ρsolid (kg/m3) at T = 298.15 K 2011 2165 2050 Elattice (kJ/mol) 786 785.5 607 TABLE VI. Densities...
2019
-
[28]
Current ad- vances in computer architecture might well remedy the com- putational overhead and make such models competitive in the very near future
This points to the limitations of point charge models and the need to address the problem of polarization in a physically meaningful way for next gener- ation force fields with enhanced transferability. Current ad- vances in computer architecture might well remedy the com- put...
2005
-
[30]
The motivation behind the choice of the JC model is three- fold
The parameters used in conjunction with the TIP4P/ε water model are shown in Table IV. The motivation behind the choice of the JC model is three- fold. Firstly, the model has been parametrized to accurately reproduce the properties of the NaCl crystal and its melt. This constr...
2005
-
[40]
Un- fortunately, the improved description of the dielectric con- stant in this model is at the cost of using the location of the hydrogen sites as fitting parameters. This can provide ex- tra flexibility for the parametrization, but will obviously ruin the molecule’s moment of...
2005
-
[58]
A number of different studies point to the robustness of this mapping to change of model po- tentials and local environmental conditions58–61 Instantaneous frequencies are time averaged over 100 ps, as recommended in Ref.62 Averages are collected from NVE simulations with 256 ...
2005
-
[72]
The num- ber of molecules and box sizes varied in different simulation sets and are summarized in Table I
Water was modeled using the TIP4P/ε force field39 and for selected properties we have also performed calculations for the TIP4P/2005 model. The num- ber of molecules and box sizes varied in different simulation sets and are summarized in Table I. Trajectories are evolved using...
2005
-
[82]
As shown in Table II, TIP4P/ ε has lower value of QT than the TIP4P/2005, explaining the shift in melting point by about 12 K. Once the melting temperature at atmospheric conditions has been evaluated, auxiliary bulk simulations atTm have been launched in the N pT ensemble to ...
2005
-
[83]
c) Relation of self- diffusion coefficient with pressure at three different temperatures for the TIP4P/ε water model
b) magnified region of part a). c) Relation of self- diffusion coefficient with pressure at three different temperatures for the TIP4P/ε water model. Experimental results is shown as a solid line. Data for T = 273 K is taken from Ref. 84 while for T = 298 K from Ref.85. By vir...
2005
-
[90]
Temperature of maximum density The TMD is quite sensitive to the addition of salts
b) Density as a function of temperature for two different concentrations for the JC-TIP4P/ ε electrolyte solu- tions.Experimental data and for Madrid-2019 force field is taken from91 C. Temperature of maximum density The TMD is quite sensitive to the addition of salts. Results...
2019
-
[1981]
Static dielectric properties of a flexible water model,
pp. 183–328, https://onlinelibrary.wiley.com/doi/pdf/10.1002/9780470142684.ch3. 44A. D. Trokhymchuk, M. F. Holovko, and K. Heinzinger, “Static dielectric properties of a flexible water model,” J. Chem. Phys. 99, 2964–2971 (1993), https://pubs.aip.org/aip/jcp/article- pdf/99/4/...
1993
-
[1991]
Determination of alkali and halide monovalent ion parameters for use in explicitly solvated biomolecular sim- ulations,
pp. 1–674. 30I. S. Joung and T. E. I. Cheatham, “Determination of alkali and halide monovalent ion parameters for use in explicitly solvated biomolecular sim- ulations,” The Journal of Physical Chemistry B 112, 9020–9041 (2008). 31T. Yagasaki, M. Matsumoto, and H. Tanaka, “Len...
2008
-
[2023]
Contratos Predoctorales de Per- sonal Investigador en Formación
CDAG acknowledges the predoctoral fellowship "Contratos Predoctorales de Per- sonal Investigador en Formación" from the Universidad Com- plutense de Madrid (CT15/23) and additional funding from Ministerio de Ciencia, Innovación y Universidades (MICIU), under grant FPU22/00869....
2016 doi
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.