Pith. sign in

REVIEW 3 major objections 4 minor 63 references

A neural network-based four-body potential energy surface for parahydrogen

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A new four-body potential energy surface for parahydrogen is net repulsive at short range and should bring dense solid hydrogen's equation of state closer to experiment.

desk verdict First ab initio four-body PES for para-H2: solid construction, honest limitations, but the short-range extrapolation and the claimed EOS improvement are not yet validated. read the letter →

arxiv 2506.05540 v1 pith:XQS3JKE7 submitted 2025-06-05 physics.chem-ph

Alexander Ibrahim , Pierre-Nicholas Roy This is my paper
classification physics.chem-ph
keywords four-bodyinteractionpotentialenergysurfaceparahydrogencoupledclustertheoryneuralnetworkequationofstatemany-bodydispersionhcplattice
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 constructs a six-dimensional, isotropic four-body interaction potential energy surface for parahydrogen, capturing the energy that appears when four H2 molecules come close together. Its central claim is that this four-body term is net repulsive at short intermolecular separations, and that adding it to the FSH pair potential and the previously computed three-body potential should place the pressure-density equation of state of solid parahydrogen closer to experiment. The surface is built from counterpoise-corrected coupled-cluster energies, orientation-averaged over molecular angles, and fitted with a multilayer perceptron that reproduces held-out test energies with a root-mean-square error of $0.066\ \mathrm{cm}^{-1}$. Why this matters: the pair-only description overestimates the pressure of dense solid hydrogen, the pair-plus-three-body description underestimates it, and the four-body term has the right sign to lie between them for most densities.

What carries the argument

The load-bearing object is the six-dimensional isotropic four-body surface $V_4(r_{12}, r_{13}, r_{14}, r_{23}, r_{24}, r_{34})$, defined on the centres of mass of four rigid para-H2 molecules. Each training energy is a counterpoise-corrected CCSD(T) interaction energy, averaged over 81 angular orientation combinations from a 6-point Lebedev quadrature. The fitting machinery is a multilayer perceptron whose input features are the reciprocals of the six side lengths, linearly rescaled to $(0,1]$, and reordered into the lexicographically smallest of the 24 allowed index permutations, which enforces invariance under swapping identical molecules. Before training, output energies are divided by the positive rescaling function $\phi(r) = A e^{-Br} + C r^{-12}$, compressing a five-order-of-magnitude energy range into roughly two orders and making weak long-range samples learnable; predictions are recovered by multiplying by $\phi(r)$. Outside the sampled region, geometries with a side length below 2.2 Å are treated with an exponential-decay extrapolation in the uniform scaling variable, while geometries with average side lengths above 4 Å are transitioned to the analytic Bade quadruple-dipole potential.

What would settle it

Run CCSD(T) calculations with an AVTZ or AVQZ basis for a tetrahedral (para-H2)4 geometry with side length 2.0 Å, and compare the four-body interaction energy with the value predicted by the paper's exponential-decay extrapolation; a deviation much larger than the neural-network RMSE would show the short-range branch is not reliable. Alternatively, perform path-integral Monte Carlo simulations of solid para-H2 at densities above $0.08\ \mathrm{\AA}^{-3}$ using the FSH pair, three-body, and four-body potentials and compare the pressure-density curve with experiment: if the curve does not move closer to experiment, the four-body surface or its extrapolation is missing relevant physics.

Watch

Extended reading notes

Core claim

The central discovery is that the four-body interaction between para-H2 molecules is predominantly repulsive at short range, the opposite sign of the net-attractive three-body interaction, and that the repulsion is quantitatively significant for dense solid hydrogen. For a tetrahedral arrangement with side length 2.2 Å, the isotropic four-body energy is roughly $200\ \mathrm{cm}^{-1}$, against a pair energy of $1079.4\ \mathrm{cm}^{-1}$ and a three-body energy of $-578.4\ \mathrm{cm}^{-1}$ for the corresponding close-packed geometries. The largest continuously differentiable neural network model reproduces the test set with an RMSE of $0.066\ \mathrm{cm}^{-1}$, and the surface merges smoothly into the analytic Bade quadruple-dipole dispersion potential at long range. In a classical frozen hcp lattice, adding the four-body term moves the pressure-density curve above the pair-plus-three-body curve below a density of about $0.08\ \mathrm{\AA}^{-3}$, so the full two-plus-three-plus-four-body description is expected to lie closer to experiment than either lower-order description alone.

Load-bearing premise

The load-bearing premise is that, for four-body geometries with side lengths below 2.2 Å, the interaction energy keeps varying exponentially as all six side lengths are uniformly scaled, with the exponential coefficient fixed by ab initio data that only reaches down to 2.25 Å; the paper itself states that this extrapolation is unlikely to be accurate for much shorter side lengths, and those short geometries carry more Boltzmann weight at the high densities where the four-body term is most important.

Editorial extensions

If this is right

  • Combined with the FSH pair potential and the isotropic three-body PES, the four-body PES should place the pressure-density equation of state of solid para-H2 between the known too-high pair-only curve and the known too-low pair-plus-three-body curve below about $0.08\ \mathrm{\AA}^{-3}$, except at very high densities where the paper expects the correction to overshoot.
  • Because the four-body term is repulsive at short range while the three-body term is attractive, the two corrections partially cancel, so omitting the four-body term either overestimates or underestimates the pressure depending on density.
  • The continuously differentiable 64-128-128-64-SSP model means the PES can be used for force evaluations, not just energies, in simulations that need gradients.
  • The long-range part is controlled by a single coefficient that can be replaced in the code without retraining the neural network, so future improvements to the four-body dispersion coefficient can be incorporated directly.
  • Because the surface is isotropic and only depends on the six centre-of-mass distances, it can be used in existing simulations alongside the FSH pair and three-body PES without introducing angular degrees of freedom.

Reading between the lines

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

  • Editorial inference: the paper's short-range extrapolation is the part most likely to be wrong in the density regime where four-body effects matter most; a direct check of the four-body energy for compact geometries with side lengths below 2.2 Å at a higher-level basis would test whether the claimed EOS improvement survives at the highest densities.
  • Editorial inference: the observed sign alternation (repulsive pair, attractive three-body, repulsive four-body) matches the pattern seen in rare-gas solids, and suggests a five-body term would again be attractive; the present data do not test this.
  • Editorial inference: the combination of permutational sorting, output rescaling, and smooth analytic switching is transferable to machine-learned many-body potentials for other weakly interacting molecular crystals, such as deuterium or HD, where high-density simulations are similarly sensitive to nonadditive forces.
  • Editorial inference: the paper's permutation-symmetry handling at the input level rather than by data augmentation implies that smaller, faster models can retain accuracy, which may be a useful design choice for other simulation-grade neural network potentials.
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 / 4 minor

Summary. The paper constructs an isotropic four-body potential energy surface (PES) for para-H2 from CCSD(T)/AVDZ+midbond electronic structure calculations. The six intramolecular distances are sampled over a range of 2.2–4.5 Å, the rotational degrees of freedom are averaged with a 6-point Lebedev quadrature, and a multilayer perceptron is trained on the resulting energies after a rescaling transformation. The PES is extended to shorter distances by an exponential-decay extrapolation and to longer distances by matching the Bade four-body dispersion potential. The authors show that the four-body interaction is net repulsive at short range, and a classical frozen-hcp calculation indicates that adding the four-body term moves the pressure-density curve between the too-repulsive pair-only result and the overcorrected pair-plus-three-body result, supporting the expectation of improved agreement with experiment.

Significance. If the PES is accurate, it is a valuable new ingredient for simulations of solid para-H2, where three-body interactions are known to overcorrect the pair-only equation of state. The paper is honest and unusually thorough in its assessment of electronic-structure errors: it reports BSSE extrapolations, compares AVDZ and AVTZ results, and gives both ab initio and independent Midzuno-Kihara estimates of the B12 coefficient. The machine-readable code and open data are strengths, and the held-out test RMSE of 0.066 cm^-1 for the 64-128-128-64-SSP model is good. However, the central claim that the PES will improve the equation of state at high densities depends on the short-range extrapolation below 2.2 Å, which the paper itself flags as unreliable, and on a classical frozen-lattice calculation rather than a quantum simulation.

major comments (3)
  1. [Sec. III C and Suppl. Secs. VI-VII] The short-range extrapolation is load-bearing for the claim that the PES will improve the solid EOS at high densities, but it is not validated in the regime where it matters most. The exponential coefficient c in Eq. (S3) is determined from ab initio energies at shortest side lengths of 2.25 and 2.26 Å and is then assumed to hold for all shorter geometries under uniform scaling. The paper explicitly states in Suppl. Sec. VII that predictions for side lengths much less than 2.2 Å are unlikely to be accurate and that this may become significant in solids at very high densities. Because solid para-H2 has large zero-point delocalization, a path-integral simulation at densities near 0.1 Å^-3 (lattice constant about 2.42 Å) can sample four-body geometries with side lengths below 2.2 Å, and the classical frozen-lattice curves in Fig. 10 never exercise this region. Please provide either additional ab initio energies for a few short geometries, a sensitivity test with an alternative extrapolation, or a restriction of the central claim to densities where the shortest sampled side length remains above about 2.2 Å.
  2. [Sec. III E, Fig. 10, and Conclusion] The expected improvement in the equation of state is inferred from a classical frozen-hcp calculation, but the abstract and conclusion state the expectation without this caveat. The classical calculation neglects zero-point motion, which is the defining feature of solid para-H2 and which changes the distribution of four-body geometries sampled by the quantum crystal. The classical pressure correction is therefore only a qualitative indication; a path-integral Monte Carlo test using the new four-body PES is needed to support the wording that the PES 'is expected to provide closer agreement with experimental results.' If such a simulation is out of scope, the claim should be explicitly limited to the classical approximation.
  3. [Sec. II F and Sec. III E] The reported test RMSE is computed on held-out samples from the random exponential-distribution geometries, not on held-out hcp-lattice geometries. The hcp geometries have special, highly correlated side-length ratios and are the ones used in the solid-state analysis; the paper notes in Sec. II F that the hcp samples are harder to fit, and the training error with hcp samples removed is lower than the total training error. Since the EOS conclusions rely on these geometries, please report the model's error on the hcp-lattice geometries, for example by leaving out a subset of the 3901 hcp samples during training and testing on them. This would show whether the 0.066 cm^-1 RMSE is representative of performance in the solid-state application.
minor comments (4)
  1. [Fig. 8 caption] The caption says 'side length 2.2 Å' for both the top and bottom panels, but the text in Sec. III B indicates the bottom panel is for 2.95 Å; please correct this.
  2. [Suppl. Sec. VI, Eq. (S5)] The linear extrapolation V_4^(li)(g) is used in Eq. (S5) but never explicitly defined; please state how it is constructed from the values at the scaling points.
  3. [Sec. III A and Table I] The Lebedev quadrature comparison is performed only for symmetric tetrahedra; a check on one or two distorted, non-tetrahedral geometries would strengthen the claim that the 6-point averaging is adequate for the general six-dimensional PES.
  4. [Sec. II C] The description of the sampling distribution says side lengths are sampled between r_min and r_max, but the specific values used (2.2 Å and 4.5 Å, with the second set starting at 2.8 Å) appear only later; it would be clearer to define r_min and r_max explicitly at the point where Eq. (4) is introduced.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the four-body PES is a direct regression on CCSD(T) data; self-citations and extrapolation limitations do not reduce the central claim to its inputs.

full rationale

The central claim is that a neural-network PES reproduces CCSD(T) four-body interaction energies and will improve solid-para-H2 EOS predictions when combined with pair and three-body potentials. This is a regression-plus-application chain, not a derivation from the claim. The NN is trained on 16,000 counterpoise-corrected CCSD(T) energies plus 1,610 hcp-lattice samples and tested on a held-out set (RMSE 0.066 cm^-1 for the 64-128-128-64-SSP model), so the accuracy claim is checked against independent ab initio data rather than defined by the model. The long-range Bade tail is an analytic adjustment; although one of the B12 estimates is obtained from the same CCSD(T) tetrahedron energies, the paper explicitly states that the neural network does not depend on B12 and that the value can be replaced without retraining, and the Midzuno-Kihara estimate is in reasonable agreement, so the tail is not forced by the fit. The short-range exponential extrapolation is a genuine limitation, not a circular step: the authors state in Supplementary Material Sec. VII that 'the exponential coefficient for the fit is determined by ab initio data that only goes down to side lengths of 2.25 Å. The short-range energy predictions made by this PES for four-body geometries with side lengths much less than 2.2 Å are unlikely to be accurate. In cases where the Boltzmann weight of such small four-body geometries is non-negligible, such as in solids at very high densities, such a limitation might become significant.' This is an unvalidated extrapolation in the regime relevant to high-density solids, so it would be a correctness/robustness concern in an application, but it does not make any prediction equivalent to an input. Self-citations to the authors' earlier three-body PES and EOS studies are used for context and for the downstream EOS comparison, not to justify the four-body PES itself; the four-body surface is built from new CCSD(T) data. No step exhibits the quoted reduction of an output to a fitted input, so no circularity is present.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests primarily on domain assumptions about the adequacy of the rigid-rotor, isotropic averaging, and CCSD(T)/AVDZ electronic structure method, plus two ad hoc extrapolation schemes (exponential decay below 2.2 Å, Bade potential above 4 Å). No new physical entities are introduced. The free parameters are the hand-chosen or fitted constants in the rescaling function, the sampling distribution, the B12 coefficient, and the transition thresholds.

free parameters (4)
  • Rescaling function parameters A, B, C = A = 3.1803e6 cm^-1, B = 4.623057 Å^-1, C = 4220.011 Å^12 cm^-1
    Used in Eq. (7) to rescale output energies during training; chosen to compress the dynamic range, not derived from first principles.
  • B12 Bade coefficient = 29492.8 cm^-1 Å^12 (AVTZ-based estimate)
    Sets the long-range four-body dispersion tail; estimated by fitting the r^-12 decay of the same type of CCSD(T) tetrahedron energies, with alternative estimates from AVDZ and Midzuno-Kihara.
  • Sampling rate constants = C0 = 0.9209 Å^-1; also 3C0, 6C0, C0/6
    Control the distribution of six side lengths in the training data (Eq. (4)); chosen by hand to emphasize short and long ranges.
  • Transition and short-range thresholds = r_L = 4 Å, r_U = 4.5 Å, R_L = 2.2 Å, R_U = 2.25 Å, c' = 6, c'' = 8, ΔR = 0.01 Å
    Hand-chosen constants for the mixing functions between PES parts and for the exponential-decay short-range extrapolation; justified by hcp lattice observations.
assumptions (6)
  • domain assumption Each para-H2 molecule is treated as a rigid rotor with a fixed bond length of 1.449 a0.
    Removes four degrees of freedom before electronic structure calculations (Sec. II B).
  • domain assumption The 6-point Lebedev quadrature with three orientations per molecule accurately approximates the full orientational average of the four-body interaction.
    Defines the isotropic V4 in Eq. (3); tested only on three tetrahedra against the 14-point scheme (Sec. III A).
  • domain assumption The many-body expansion (Eq. (1)) converges, so that pair plus three-body plus four-body potentials capture the relevant physics of solid para-H2.
    Justifies combining the new 4B PES with the FSH pair and the group's 3B PES in the EOS analysis (Sec. III E).
  • domain assumption The long-range four-body interaction is described by the quadruple-dipole Bade potential with a single B12 coefficient.
    Used to patch the PES for average side lengths above 4 Å (Sec. III D).
  • ad hoc to paper For geometries with a shortest side length below 2.2 Å, the four-body energy varies exponentially under a uniform scaling of all side lengths.
    Basis for the short-range extrapolation in Supp. Sec. VI; Supp. Sec. VII states the extrapolation is unlikely to be accurate for much shorter distances.
  • domain assumption A classical frozen hcp lattice calculation is indicative of how the four-body potential will affect the quantum solid para-H2 equation of state.
    Used in Sec. III E to argue the 4B PES will bring the EOS closer to experiment; no quantum PIMC simulation with the 4B PES is performed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A neural network-based four-body potential energy surface for parahydrogen." pith.science (2026). https://pith.science/paper/XQS3JKE7

@misc{pith2026250605540,
  author       = {Pith},
  title        = {Pith review of: A neural network-based four-body potential energy surface for parahydrogen},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XQS3JKE7}},
  note         = {Machine review of arXiv:2506.05540}
}
abstract

We present an isotropic ab initio (para-H$_2$)$_4$ four-body interaction potential energy surface (PES). The electronic structure calculations are performed at the correlated coupled-cluster theory level, with single, double, and perturbative triple excitations. They use an atom-centred augmented correlation-consistent double zeta basis set, supplemented by a $(3s3p2d)$ midbond function. We use a multilayer perceptron to construct the PES. We apply a rescaling transformation to the output energies during training to improve the prediction of weaker energies in the sample data. At long distances, the interaction energies are adjusted to match the empirically-derived four-body dispersion interaction. The four-body interaction energy at short intermolecular separations is net repulsive. The use of this four-body PES, in combination with a first principles pair potential for para-H$_2$ [J. Chem. Phys. 119, 12551 (2015)], and an isotropic ab initio three-body potential for para-H$_2$ [J. Chem. Phys. 156, 044301 (2022)], is expected to provide closer agreement with experimental results.

Figures

Figures reproduced from arXiv: 2506.05540 by the authors.

Figure 1
Figure 1. FIG. 1. A graphical representation of how fixing the bond [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Normalized histograms of the sample outputs, after [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4. The base-10 logarithm of the mean squared error [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (10 more)
Figure 6
Figure 6. Figure 6: FIG. 6. The four-body interaction potential energies of a [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. The relative error of calculating the four-body in [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. The four-body interaction energy of a tetrahedron [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 6
Figure 6. Figure 6: In fact, it is the case for all 8 of the highest en￾ergy hcp geometries shown in [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 9
Figure 9. Figure 9: FIG. 9. The total contribution of each geometry to the average four-body [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10. We show the classical energy-density curves (top) [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 1
Figure 1. Figure 1: FIG. 1. The rescaled four-body interaction energy for a tetrahedron geometry, as a function of the tetrahedron’s side length [PITH_FULL_IMAGE:figures/full_fig_p018_1.png]
Figure 2
Figure 2. Figure 2: FIG. 2. The base-10 logarithm of the mean squared error loss as a function of the epoch during the training of the [PITH_FULL_IMAGE:figures/full_fig_p019_2.png]
Figure 3
Figure 3. Figure 3: FIG. 3. The base-10 logarithm of the mean squared error loss as a function of the epoch during the training of the [PITH_FULL_IMAGE:figures/full_fig_p019_3.png]
Figure 4
Figure 4. Figure 4: FIG. 4. The base-10 logarithm of the mean squared error loss as a function of the epoch during the training of the [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 31 canonical work pages

  1. [1]

    G. R. Medders, A. W. G \"o tz, M. A. Morales, P. Bajaj, and F. Paesani, ``On the representation of many-body interactions in water,'' J.\ Chem.\ Phys. , vol. 143, p. 104102, 2015

    work page 2015

  2. [2]

    Z. Fan, W. Chen, V. Vierimaa, and A. Harju, ``Efficient molecular dynamics simulations with many-body potentials on graphics processing units,'' Comput.\ Phys.\ Commun. , vol. 218, pp. 10--16, 2017

    work page 2017

  3. [3]

    Alkan, P

    M. Alkan, P. Xu, and M. S. Gordon, ``Many-body dispersion in molecular clusters,'' J.\ Phys.\ Chem.\ A , vol. 123, pp. 8406--8416, 2019

    work page 2019

  4. [4]

    A. O. de-la Roza, L. M. LeBlanc, and E. R. Johnson, ``What is ``many-body'' dispersion and should I worry about it?,'' Phys.\ Chem.\ Chem.\ Phys. , vol. 22, pp. 8266--8276, 2020

    2020

  5. [5]

    M. J. Elrod and R. J. Saykally, ``Many-body effects in intermolecular forces,'' Chem.\ Rev. , vol. 94, p. 1975, 1994

    1975

  6. [6]

    G. A. Cisneros, K. T. Wikfeldt, L. Ojam \"a e, J. Lu, Y. Xu, H. Torabifard, A. P. Bart \'o k, G. Cs \'a nyi, V. Molinero, and F. Paesani, ``Modeling molecular interactions in water: From pairwise to many-body potential energy functions,'' Chem.\ Rev. , vol. 116, pp. 7501--7528, 2016

    work page 2016

  7. [7]

    H. P. Gush, W. F. J. Hare, E. J. Allin, and H. L. Welsh, ``The infrared fundamental band of liquid and solid hydrogen,'' Can.\ J.\ Phys. , vol. 38, p. 176, 1960

    1960

  8. [8]

    L. H. Nosanow, ``Theory of quantum crystals,'' Phys.\ Rev. , vol. 146, p. 120, 1966

    1966

Show all 63 references
  1. [9]

    Bostanjoglo and R

    O. Bostanjoglo and R. Kleinschmidt, ``Crystal structure of hydrogen isotopes,'' J.\ Chem.\ Phys. , vol. 46, p. 2004, 1967

    2004

  2. [10]

    van Kranendonk, Solid hydrogen: T heory of the properties of solid H _2 , H D , and D _2

    J. van Kranendonk, Solid hydrogen: T heory of the properties of solid H _2 , H D , and D _2 . Boston, MA: Springer, 1983

    1983

  3. [11]

    Fernandez-Alonso, C

    F. Fernandez-Alonso, C. Cabrillo, R. Fern\' a ndez-Perea, F. J. Bermejo, M. A. Gonz\' a lez, C. Mondelli, and E. Farhi, ``Solid para-hydrogen as the paradigmatic quantum crystal: T hree observables probed by ultrahigh-resolution neutron spectroscopy,'' Phys.\ Rev.\ B , vol. 86...

    2012

  4. [12]

    Dusseault and M

    M. Dusseault and M. Boninsegni, ``Atomic displacements in quantum crystals,'' Phys.\ Rev.\ B , vol. 95, p. 104518, 2017

    2017

  5. [13]

    I. F. Silvera, ``The solid molecular hydrogens in the condensed phase: F undamentals and static properties,'' Rev.\ Mod.\ Phys. , vol. 52, p. 393, 1980

    1980

  6. [14]

    Zoppi and M

    M. Zoppi and M. Neumann, `` PIMC simulations of solid parahydrogen,'' Phys.\ Rev.\ B , vol. 43, no. 13, p. 10242, 1991

    1991

  7. [15]

    D. M. Ceperley, ``Path integrals in the theory of condensed helium,'' Rev.\ Mod.\ Phys. , vol. 67, p. 279, 1995

    1995

  8. [16]

    Wind and I

    P. Wind and I. R eggen, ``Ab initio calculation of three-body interaction in the ( H _2 ) _3 trimer,'' Chem.\ Phys. , vol. 211, pp. 179--189, 1996

    1996

  9. [17]

    R. J. Hinde, ``Three-body interactions in solid parahydrogen,'' Chem.\ Phys.\ Lett. , vol. 460, pp. 141--145, 2008

    2008

  10. [18]

    Manzhos, K

    S. Manzhos, K. Nakai, and K. Yamashita, ``Three-body interactions in clusters C O --(p H _2 ) _n ,'' Chem.\ Phys.\ Lett. , vol. 493, pp. 229--233, 2010

    2010

  11. [19]

    Ibrahim and P.-N

    A. Ibrahim and P.-N. Roy, ``Three-body potential energy surface for para-hydrogen,'' J.\ Chem.\ Phys. , vol. 156, p. 044301, 2022

    2022

  12. [20]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, ``Pytorch: an imperative style, high-p...

    2019

  13. [21]

    Abadi, A

    M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Goodfellow, A. Harp, G. Irving, M. Isard, Y. Jia, R. Jozefowicz, L. Kaiser, M. Kudlur, J. Levenberg, D. Man \'e , R. Monga, S. Moore, D. Murray, C. Olah, ...

    2015

  14. [22]

    J. O. Hirschfelder, R. B. Bird, and E. L. Spotz, ``The transport properties for non-polar gases,'' J.\ Chem.\ Phys. , vol. 16, no. 10, p. 968, 1948

    1948

  15. [23]

    E. A. Mason and W. E. Rice, ``The intermolecular potentials of helium and hydrogen,'' J.\ Chem.\ Phys. , vol. 22, no. 5, p. 522, 1954

    1954

  16. [24]

    J. M. Farrer and Y. T. Lee, ``Intermolecular potentials from crossed beam differential elastic scattering measurements. V I I para- H _2 + para- H _2 ,'' J.\ Chem.\ Phys. , vol. 57, no. 12, p. 5492, 1972

    1972

  17. [25]

    I. F. Silvera and V. V. Goldman, ``The isotropic intermolecular potential for H _2 and D _2 in the solid and gas phases,'' J.\ Chem.\ Phys. , vol. 69, p. 4209, 1978

    1978

  18. [26]

    U. Buck, F. Huisken, A. Kohlhase, D. Otten, and J. Schaefer, ``State resolved rotational excitation in D _2 + H _2 collisions,'' J.\ Chem.\ Phys. , vol. 78, p. 4439, 1978

    1978

  19. [27]

    M. J. Norman, R. O. Watts, and U. Buck, ``A spherical potential for hydrogen from solid state and scattering data,'' J.\ Chem.\ Phys. , vol. 81, p. 3500, 1984

    1984

  20. [28]

    Diep and J

    P. Diep and J. K. Johnson, ``An accurate H _2 -- H _2 interaction potential from first principles,'' J.\ Chem.\ Phys. , vol. 112, no. 10, p. 4465, 2000

    2000

  21. [29]

    R. J. Hinde, ``A six-dimensional H _2 -- H _2 potential energy surface for bound state spectroscopy,'' J.\ Chem.\ Phys. , vol. 128, p. 154308, 2008

    2008

  22. [30]

    Patkowski, W

    K. Patkowski, W. Cencek, P. Jankowski, K. Szalewicz, J. B. Mehl, G. Garberoglio, and A. H. Harvey, ``Potential energy surface for interactions between two hydrogen molecules,'' J.\ Chem.\ Phys. , vol. 129, p. 094304, 2008

    2008

  23. [31]

    Faruk, M

    N. Faruk, M. Schmidt, H. Li, R. J. Le Roy, and P.-N. Roy, ``First-principles prediction of the R aman shifts in parahydrogen clusters,'' J.\ Chem.\ Phys. , vol. 141, p. 014310, 2014

    2014

  24. [32]

    Omiyinka and M

    T. Omiyinka and M. Boninsegni, ``Pair potentials and equation of state of solid para-hydrogen to megabar pressure,'' Phys.\ Rev.\ B , vol. 88, p. 024112, 2013

    2013

  25. [33]

    Operetto and F

    F. Operetto and F. Pederiva, ``Diffusion M onte C arlo study of the equation of state of solid para- H _2 ,'' Phys.\ Rev.\ B , vol. 73, p. 184124, 2006

    2006

  26. [34]

    B. M. Axilrod and E. Teller, ``Interaction of the van der W aals type between three atoms,'' J.\ Chem.\ Phys. , vol. 11, p. 299, 1943

    1943

  27. [35]

    Muto, ``Force between nonpolar molecules,'' J.\ Phys.\ Math.\ Soc.\ Jpn

    Y. Muto, ``Force between nonpolar molecules,'' J.\ Phys.\ Math.\ Soc.\ Jpn. , vol. 17, p. 629, 1943

    1943

  28. [36]

    Tao and Y.-K

    F.-M. Tao and Y.-K. Pan, ``M ller- P lesset perturbation investigation of the H e _2 potential and the role of midbond basis functions,'' J.\ Chem.\ Phys. , vol. 97, no. 7, p. 4989, 1992

    1992

  29. [37]

    Hollebeek, T.-S

    T. Hollebeek, T.-S. Ho, H. Rabitz, and L. B. Harding, ``Construction of reproducing kernel H ilbert space potential energy surfaces for the 1 A '' and 1 A ' states of the reaction N (^2 D ) + H _2 ,'' J.\ Chem.\ Phys. , vol. 114, p. 3945, 2001

    2001

  30. [38]

    Ho and H

    T.-S. Ho and H. Rabitz, ``Reproducing kernel H ilbert space interpolation methods as a paradigm of high dimensional model representations: A pplication to multidimensional potential energy surface construction,'' J.\ Chem.\ Phys. , vol. 119, no. 13, p. 6433, 2003

    2003

  31. [39]

    O. T. Unke and M. Meuwly, ``Toolkit for the construction of reproducing kernel-based representations of data: A pplication to multidimensional potential energy surfaces,'' J.\ Chem.\ Inf.\ Model. , vol. 57, p. 1923, 2017

    1923

  32. [40]

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

    O. T. Unke, S. Chmiela, H. E. Sauceda, M. Gastegger, I. Poltavsky, K. T. Sch \"u tt, A. Tkatchenko, and K.-R. M \"u ller, ``Machine learning force fields,'' Chem.\ Rev. , vol. 121, pp. 10142--10186, 2021

    2021

  33. [41]

    Ibrahim and P.-N

    A. Ibrahim and P.-N. Roy, ``Equation of state of solid parahydrogen using ab initio two-body and three-body interaction potentials,'' J.\ Chem.\ Phys. , vol. 157, p. 174503, 2022

    2022

  34. [42]

    Ibrahim, L

    A. Ibrahim, L. Wang, T. Halverson, R. J. Le Roy , and P.-N. Roy, ``Equation of state and first principles prediction of the vibrational matrix shift of solid parahydrogen,'' J.\ Chem.\ Phys. , vol. 151, p. 244501, 2019

    2019

  35. [43]

    R. J. Wheatley, G. Garberoglio, and A. H. Harvey, ``Four-body nonadditive potential energy surface and the fourth virial coefficient of helium,'' J.\ Chem.\ Eng.\ Data. , vol. 68, pp. 3257--3264, 2023

    2023

  36. [44]

    R. S. Graham and R. J. Wheatley, ``Machine learning for non-additive intermolecular potentials: Q uantum chemistry to first-principles predictions,'' Chem.\ Commun. , vol. 58, p. 6898, 2022

    2022

  37. [45]

    V. I. Lebedev, ``Quadratures on a sphere,'' Zh.\ Vychisl.\ Mat.\ Mat.\ Fiz. , vol. 16, no. 2, pp. 293--306, 1976

    1976

  38. [46]

    W. L. Bade, ``Drude-model calculation of dispersion forces. I . G eneral theory,'' J.\ Chem.\ Phys. , vol. 27, pp. 1280--1284, 1957

    1957

  39. [47]

    W. L. Bade, ``Drude-model calculation of dispersion forces. III . T he fourth-order contribution,'' J.\ Chem.\ Phys. , vol. 28, pp. 282--284, 1958

    1958

  40. [48]

    Rolik, L

    Z. Rolik, L. Szegedy, I. Ladj\' a nszki, B. Lad\' o czki, and M. K\' a llay, ``An efficient linear-scaling C C S D ( T ) method based on local natural orbitals,'' J.\ Chem.\ Phys. , vol. 139, p. 094105, 2013

    2013

  41. [49]

    Raghavachari, G

    K. Raghavachari, G. W. Trucks, J. A. Pople, and M. Head-Gordon, ``A fifth-order perturbation comparison of electron correlation theories,'' Chem.\ Phys.\ Lett. , vol. 157, no. 6, pp. 479--483, 1989

    1989

  42. [50]

    S. F. Boys and F. Bernardi, ``The calculation of small molecular interactions by the differences of separate total energies. S ome procedures with reduced errors,'' Mol.\ Phys. , vol. 19, no. 4, p. 553, 1970

    1970

  43. [51]

    A. D. Becke, ``A multicenter numerical integration scheme for polyatomic molecules,'' J.\ Chem.\ Phys. , vol. 88, no. 4, p. 2547, 1988

    1988

  44. [52]

    Wang and T

    X.-G. Wang and T. Carrington . Jr., ``Using L ebedev grids, sine spherical harmonics, and monomer contracted basis functions to calculate being energy levels of H F trimer,'' J. Theor.\ Comp.\ Chem. , vol. 2, no. 4, pp. 599--608, 2003

    2003

  45. [53]

    Garberoglio, ``On the contribution of non-additive three-body interactions to the third virial coefficient of para-hydrogen,'' Chem.\ Phys.\ Lett

    G. Garberoglio, ``On the contribution of non-additive three-body interactions to the third virial coefficient of para-hydrogen,'' Chem.\ Phys.\ Lett. , vol. 557, p. 26, 2012

    2012

  46. [54]

    B. T. Fang, ``Trilateration and extension to global positioning system navigation,'' J.\ Guid.\ Control\ Dyn. , vol. 9, pp. 715--717, 1986

    1986

  47. [55]

    Fukushima, ``Cognitron: A self-organizing multilayered neural network,'' Biol.\ Cybernetics , vol

    K. Fukushima, ``Cognitron: A self-organizing multilayered neural network,'' Biol.\ Cybernetics , vol. 20, pp. 121--136, 1975

    1975

  48. [56]

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

    K. T. Sch \"u tt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, and K.-R. M \"u ller, ``Sch N et: A deep learning architecture for molecules and materials,'' J.\ Chem.\ Phys. , vol. 148, p. 241722, 2018

    2018

  49. [57]

    LeNail, `` N N - S V G : P ublication-ready neural network architecture schematics,'' J.\ Open\ Source\ Softw

    A. LeNail, `` N N - S V G : P ublication-ready neural network architecture schematics,'' J.\ Open\ Source\ Softw. , vol. 4, p. 747, 2019

    2019

  50. [58]

    Midzuno and T

    Y. Midzuno and T. Kihara, ``Non-additive intermolecular potential in gases I . v an der W aals interactions,'' J.\ Phys.\ Soc.\ Jpn. , vol. 11, no. 10, pp. 1045--1049, 1956

    1956

  51. [59]

    Schmidt, J

    M. Schmidt, J. M. Fern\' a ndez, N. Faruk, M. Nooijen, R. J. Le Roy , J. H. Morilla, G. Tejeda, S. Montero, and P.-N. Roy, ``Raman vibrational shifts of small clusters of hydrogen isotopologues,'' J.\ Phys.\ Chem.\ A , vol. 119, p. 12551, 2015

    2015

  52. [60]

    C. P. Herrero and R. Ramirez, ``Path-integral simulation of solids,'' J.\ Phys.\ Condens.\ Matter , vol. 26, p. 233201, 2014

    2014

  53. [61]

    Tian, F.-S

    C.-L. Tian, F.-S. Liu, F.-Q. Jing, and L.-C. Cai, ``Five- and six-body effects on equation of state of solid ^4 H e,'' J.\ Phys.\ Condens.\ Matter , vol. 18, pp. 8103--8112, 2006

    2006

  54. [62]

    C.-L. Tian, N. Wu, F. Liu, S. K. Saxena, and X. Zheng, ``Four-body interaction energy for compressed solid krypton from quantum theory,'' J.\ Chem.\ Phys. , vol. 137, p. 044108, 2012

    2012

  55. [63]

    Ro\' s ciszewski, B

    K. Ro\' s ciszewski, B. Paulus, P. Fulde, and H. Stoll, ``Ab initio coupled-cluster calculations for the fcc and hcp structures of rare-gas solids,'' Phys.\ Rev.\ B , vol. 62, no. 9, p. 5482, 2000

    2000

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.