Chemical potentials from structure factors: I. Neutral multi-component mixtures

Bingqing Cheng; Musahid Ahmed; Roya Savoj; Xiaoyu Wang

REVIEW 3 major objections 5 minor 53 references

Chemical potentials from structure factors: I. Neutral multi-component mixtures

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Structure factors alone give chemical potentials for neutral mixtures.

desk verdict Solid extension of the S0 method to multicomponent neutral mixtures, with real validation; the OZ extrapolation is the soft spot but not a fatal one. read the letter →

arxiv 2608.08357 v1 pith:T3LIEP2T submitted 2026-08-08 physics.chem-ph cond-mat.softcond-mat.stat-mechphysics.comp-ph

Roya Savoj , Xiaoyu Wang , Musahid Ahmed , Bingqing Cheng This is my paper · ORCID

classification physics.chem-phcond-mat.softcond-mat.stat-mechphysics.comp-ph

keywords chemicalpotentialstructurefactorsmulti-componentmixturesS0methodGaussianprocessregressionactivelearningsolubilityOrnstein-Zernikeextrapolation

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 extends the S0 method, which reads chemical-potential derivatives off particle-number fluctuations in equilibrium simulations, from binary solutions to mixtures with any number of neutral components. The authors derive the high-dimensional version of the fluctuation-response relations, replace the scalar Ornstein–Zernike extrapolation with a matrix fit of the whole structure-factor matrix, and integrate the resulting derivative matrix with a gradient-aware Gaussian process that actively chooses which new compositions to simulate. They validate the method on a molten Fe-Cu-Ni alloy and use it to compute paracetamol solubilities in water-ethanol, matching the experimentally observed ethanol-dependent trend. If correct, the method gives a practical route to mixing free energies and solubilities that avoids particle insertion, thermodynamic integration, and real-space Kirkwood-Buff integrals.

What carries the argument

The carrying object is the matrix Ornstein–Zernike extrapolation of Eq. (8): $S_{\mathrm{OZ}}(k) = (S^{-1}(0)+k^2 L)^{-1}$, fitted simultaneously to all diagonal and off-diagonal elements of the partial structure-factor matrix. Extrapolating to $k=0$ gives $S^0$, which equals the grand-canonical fluctuation matrix $B$ scaled by $1/\sqrt{x_\alpha x_\beta}$; inverting and projecting that matrix through Eqs. (10), (15), and (18) yields the chemical-potential derivative matrix $\Gamma^{\mathrm{ex}}$ defined on the independent composition variables. The second carrying mechanism is Gaussian process regression that consumes both the derivative rows $\Gamma[\alpha,:]$ and a reference chemical potential, producing global chemical-potential surfaces with uncertainty estimates, and a CUR low-rank selection of the gradient covariance matrix that drives the active-learning loop.

What would settle it

Run the identical Fe-Cu-Ni or paracetamol systems in much larger simulation boxes (or with multiple $k_{cut}$ values) and check whether the extrapolated $S^0$ matrix is independent of box size and cutoff; if $S^0$ drifts, the chemical potentials inherit that drift. A sharper test is to compute chemical potentials at a handful of off-reference compositions with an independent method, such as free-energy perturbation or thermodynamic integration, and compare with the GP-integrated values.

Watch

Extended reading notes

Core claim

The central claim is that the matrix of chemical-potential derivatives with respect to independent mole fractions, $\Gamma = U M$, can be obtained directly from partial static structure factors $S_{\alpha\beta}(k)$ computed in ordinary NPT simulations, and that integrating $\Gamma$ yields excess chemical potentials over the whole composition space. The key step is the generalized matrix Ornstein–Zernike fit $S_{\mathrm{OZ}}(k) = (S^{-1}(0)+k^2 L)^{-1}$, which extracts the zero-wavevector matrix $S^0$ without the finite-size contamination that plagues real-space Kirkwood-Buff integrals. A Gaussian process trained on both function values and gradient data performs the integration, and CUR-based active learning chooses the most informative compositions to simulate next. The paper demonstrates the scheme on a ternary alloy, where mixing free energies agree with a CALPHAD-style thermodynamic model, and on paracetamol solubility, where the computed ethanol dependence matches experiment.

Load-bearing premise

The zero-wavevector limit of the NPT partial structure factors, obtained from the matrix Ornstein–Zernike fit with the chosen cutoff, must equal the true grand-canonical fluctuation matrix; any bias in that extrapolation propagates into every chemical potential and solubility.

Editorial extensions

If this is right

For a neutral mixture with $C$ components, only $C-1$ independent mole fractions are needed, and the formalism handles them through a projection matrix, so the method extends beyond three components.
Mixing free energies of fully miscible non-ideal ternary alloys can be computed from modest NPT simulations, as shown by the Fe-Cu-Ni validation against a CALPHAD-style thermodynamic model.
Solubilities of molecular crystals in mixed solvents follow from the intersection of the GP-integrated solute chemical potential with the solid-state chemical potential; the paracetamol example reproduces the experimental rise and plateau with ethanol content.
The combination of GP integration and CUR-based active learning lowers the number of compositions that must be simulated compared with random sampling, at fixed accuracy of the mixing free energy.
The paper explicitly limits itself to neutral components; extending to charged species needs a charge-aware small-k treatment, which the authors say will be the subject of a follow-up.

Reading between the lines

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

A systematic convergence test across box sizes and $k_{cut}$ values would directly probe the one assumption this method leans on, since the paper uses a single cutoff for the matrix Ornstein–Zernike fit.
The same derivative matrix $\Gamma$ could yield activity coefficients, osmotic compressibilities, or Kirkwood-Buff integrals over the entire composition space at no extra simulation cost.
For mixtures with more than three components, the GP integration's cost grows with dimension; the paper notes more simulations are needed, and warped or non-stationary kernels may matter more as composition space grows.

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.

Desk Editor's Note

read the letter

Quick take: this is a genuine, useful generalization of the S0 method to arbitrary multicomponent neutral mixtures, backed by two validation tests and shipped code/data. The thermodynamic derivation in Section II and the appendix is standard Kirkwood-Buff ensemble switching, and the matrix OZ fit plus GP gradient integration with CUR active learning is a sensible toolbox. The Fe-Cu-Ni mixing free energies agree well with an independent CALPHAD/Widom model, and the paracetamol solubility in pure ethanol lands inside the experimental spread. That is real evidence the method works.

The GP-with-gradients integration and the CUR-based active learning are the genuinely new bits. The C-component generalization of the S0 relation is algebraically natural, but the matrix OZ extrapolation and the GP machinery make it practical beyond binaries.

Soft spots, in order of importance. The zero-wavevector extrapolation, Eq. (8), uses a hand-chosen cutoff k_cut^2 = 0.005 × 4π^2/Å^2 with no sensitivity analysis. The stress-test note is right that the Fe-Cu-Ni test is a high-temperature metal whose structure factors are smooth, while the paracetamol system is hydrogen-bonded and molecular. That said, the pure-ethanol solubility matching experiment is a nontrivial check that the extrapolation is not badly biased in a molecular solvent. Still, I'd want a referee to ask for a cutoff sweep or a comparison with the scalar OZ form to see how much the S0 values change.

Second, the GP hyperparameters for the paracetamol system are set by hand (alpha, theta, sigma_g). The paper says these were chosen based on the underlying problem, but there's no sensitivity analysis or marginal-likelihood optimization for that system. This is minor because the Fe-Cu-Ni GP optimizes its hyperparameters and the paracetamol result is consistent with experiment, but it leaves a small question about reproducibility.

Third, the GP reconstruction does not explicitly enforce Gibbs-Duhem on the integrated surface. The derivative matrix U satisfies x^T U = 0 by construction at each point, and the GP uses those derivatives, so the violation would be small, but the paper doesn't quantify it.

There are no critical red flags. The central claim holds: chemical potentials in multicomponent neutral mixtures can be obtained from moderate equilibrium NPT runs. This is a paper for practitioners of free-energy calculations, Kirkwood-Buff users, and people doing solubility prediction. It deserves a serious referee and, given the code and data, should be publishable after minor revision addressing the cutoff sensitivity and GP hyperparameter choice.

Referee Report

3 major / 5 minor

Summary. The manuscript generalizes the S0 method to neutral multi-component mixtures. It derives the multi-component fluctuation relation between the zero-wavevector limit of partial structure factors and chemical potential derivatives, introduces a matrix Ornstein-Zernike extrapolation, and combines these ingredients with Gaussian-process regression using gradient observations and active learning. The method is applied to compute mixing free energies in liquid Fe-Cu-Ni and paracetamol solubilities in water-ethanol. The Fe-Cu-Ni results agree with an independent CALPHAD-style thermodynamic model and with the two-component S0 method; the paracetamol solubility in pure ethanol falls within the experimental spread, while the pure-water solubility is underpredicted and is attributed to force-field error.

Significance. If the S0-to-chemical-potential pipeline is reliable, the method offers a practical route to multi-component chemical potentials from equilibrium NPT simulations without particle insertion or multi-stage thermodynamic integration. The thermodynamic derivation in Section II and Appendix VI A is clean and follows standard Kirkwood-Buff theory; the Gaussian-process integration with gradient data and the active-learning selection are useful methodological advances. The manuscript provides public code and data, which is a strength. There is no circularity in the central pipeline: the derivative matrix is computed from fluctuation data, and the FEP/TI references fix only integration constants. The main uncertainty is the fidelity of the zero-wavevector extrapolation, and the validation for a high-temperature metallic liquid does not by itself guarantee the same accuracy for hydrogen-bonded molecular solutions.

major comments (3)

[Section II A, Eq. (8); Sections III and IV] The entire method rests on the identification of the extrapolated S0 with the grand-canonical fluctuation matrix B, but the only bridge from finite-k NPT structure factors to k=0 is the truncated matrix Ornstein-Zernike form in Eq. (8), used with a fixed cutoff k_cut^2 = 0.005 x 4*pi^2/Angstrom^2 for both systems. No sensitivity analysis is reported for the fitting range, the initial guess for L, or the functional form (e.g., adding a k^4 term). Because a biased S0 propagates through Eqs. (10), (15), and (18) into every chemical potential and solubility, the authors should demonstrate that S0 and the final free energies/solubilities are robust to these choices, and should include the OZ fitting uncertainty in the reported error bars.
[Section III, Fig. 3a] The RMSE comparison between CUR and random selection is evaluated against a 231-point full grid that is generated by the same S0/GP pipeline, so it measures internal consistency and sampling efficiency rather than absolute accuracy. The absolute validation is against the independent CALPHAD-style model in Fig. 4, but only for fixed x_Cu slices. The manuscript should state this distinction explicitly, and ideally report a single quantitative error metric against the thermodynamic model over the full composition range used.
[Section IV, Fig. 6b] The experimental validation for the mixed-solvent solubility curve is largely qualitative: only the pure-ethanol Form I solubility is matched within the experimental scatter, while the pure-water solubility is below the experimental range. The force-field explanation is plausible but is not quantified here. Since the mixed-solvent trend is the paper's main molecular demonstration, a quantitative comparison against the three experimental datasets, or an independent FEP/TI point at an intermediate solvent composition, would materially strengthen the claim that the method captures the solvent-composition dependence.

minor comments (5)

[Section II A, Eq. (5)] The NPT structure factor at fixed particle numbers has a trivial k=0 value, so the equivalence in Eq. (4) is a grand-canonical statement that relies on the thermodynamic limit and on the OZ extrapolation; a sentence stating this explicitly where Eq. (5) is introduced would help avoid confusion.
[Section IV] The choice of representative atoms (oxygen, nitrogen, hydroxyl-adjacent carbon) rather than molecular centers of mass should be justified; the k=0 limit is the same, but the finite-k behavior and hence the OZ fit may differ.
[Section II E and Fig. 3] The mapping between active-learning rounds and the number of selected compositions is not stated; please report the number of points added per round, m, and the stopping criterion used in the CUR selection.
[Section IV, Eqs. (23)-(24)] The GP hyperparameters alpha, theta, and sigma_g are quoted as fixed values, but it is not stated whether they were optimized, chosen by cross-validation, or set a priori; please clarify.
[Section IV] At the lowest paracetamol mole fraction (x_para = 5e-4), the number of paracetamol molecules in a box of roughly 10,000 solvent molecules is small; the statistical reliability of the S_para-para and S_para-solvent structure factors in this regime should be discussed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: chemical potentials are integrated from structure-factor fluctuation data, with independent FEP/TI integration constants and external CALPHAD/experimental benchmarks.

full rationale

The derivation chain is self-contained and non-circular. The paper computes S0 by fitting the matrix Ornstein–Zernike form (Eq. 8) to finite-k structure factors from NPT MD, then uses the exact grand-canonical fluctuation relation (Eq. 4) and the ensemble-switching identity (Eq. 10, derived in Appendix VI A from Gibbs–Duhem and Maxwell relations) to build U, projects to independent mole fractions (Eqs. 11–15), and integrates the resulting gradients with a Gaussian process (Eqs. 19–23). The only fitted quantities—S0 and L in the OZ fit, and GP hyperparameters—are fitted to structure-factor data and gradient observations, not to the target solubility or mixing free energy. The absolute chemical potentials used as GP integration constants are taken from Ref. [16]'s FEP/TI calculations in pure solvents at finite paracetamol mole fractions; these are independent calculations of a different quantity (absolute excess chemical potential), not the solubility itself, and the concentration-dependent slope that determines solubility comes from the S0 gradients. The solid chemical potentials are similarly independent TI results. The Fe-Cu-Ni validation against the CALPHAD-style model of Ref. [38] and the paracetamol comparison with experimental solubility data and FEP/BAR estimates provide external benchmarks. No equation reduces by construction to its inputs, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The derivation rests on standard fluctuation-dissipation relations, on the equivalence between NPT structure factors and grand canonical fluctuations in the thermodynamic limit, on the truncated Ornstein-Zernike form used to extrapolate S0, and on the accuracy of the chosen force fields and GP hyperparameters. No new physical entities are introduced.

free parameters (4)

Ornstein-Zernike inverse-slope matrix L = per composition, fitted to S(k)
The matrix OZ form in Eq. (8) uses L as a fitting parameter to extrapolate S(k) to k = 0; the resulting S0 values feed directly into Eqs. (4), (10), and (15), so the chemical potentials depend on this fit.
OZ extrapolation cutoff k_cut^2 = 0.005 x 4 pi^2 / Angstrom^2 for both systems
Chosen by hand in Sections III and IV; it controls the wavevector range used for the S0 extrapolation and could bias the extrapolated values if the OZ form is not valid over that range.
GP length scales theta and gradient noise sigma_g = Fe-Cu-Ni: optimized by marginal likelihood; paracetamol: theta = (0.12, 0.13), sigma_g = 0.10 kJ/mol
Hyperparameters of the Gaussian process integration; the paracetamol values were set manually, not fitted to solubility, but they affect the reconstructed chemical potential surface and its uncertainty.
Input warping parameter alpha for paracetamol GP = alpha = (0, 0.125)
This non-stationary kernel warp in Eq. (24) is an additional hand-chosen parameter that changes the effective length scale along the ethanol dimension, and therefore the GP interpolation.

assumptions (5)

standard math The fluctuation-dissipation relation in Eq. (2) correctly links particle-number fluctuations in the grand canonical ensemble to chemical potential derivatives.
This is the statistical mechanical identity on which the entire S0 method rests, stated in Section II A.
domain assumption The k-to-zero limit of partial structure factors computed from finite-size NPT simulations equals the grand canonical fluctuation matrix in the thermodynamic limit.
The method uses S(k) from NPT MD simulations at fixed particle numbers and extrapolates to k = 0; this requires ensemble equivalence and is implicitly assumed in Section II A.
ad hoc to paper The matrix Ornstein-Zernike form in Eq. (8), truncated at k^2, is a valid representation of the small-wavevector behavior of S(k) for the neutral mixtures studied.
This is a modeling choice introduced to extrapolate S0; its validity is not proven and the cutoff is chosen by hand in Sections III and IV.
domain assumption The force fields used (EAM for Fe-Cu-Ni, CHARMM36 for the paracetamol system) are accurate enough for the target thermodynamic quantities.
The paper itself attributes the poor paracetamol solubility in pure water to force-field limitations, so the accuracy of the results is conditional on the interatomic potentials.
ad hoc to paper The Gaussian process kernel and its hyperparameters provide a faithful model of the chemical potential surface over the composition space.
The GP integration in Section II D assumes a smoothness prior that is not derived from the physics; the paracetamol hyperparameters are set manually, and the Fe-Cu-Ni ones are optimized on the same gradient data used for prediction.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Chemical potentials from structure factors: I. Neutral multi-component mixtures." pith.science (2026). https://pith.science/paper/T3LIEP2T

@misc{pith2026260808357,
  author       = {Pith},
  title        = {Pith review of: Chemical potentials from structure factors: I. Neutral multi-component mixtures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T3LIEP2T}},
  note         = {Machine review of arXiv:2608.08357}
}

read the original abstract

The chemical potentials of multi-component mixtures underlie many physical and chemical phenomena, but remain challenging to compute. The S0 method enables the computation of chemical potentials from equilibrium molecular dynamics simulations, by leveraging the thermodynamic relationship between particle number fluctuations and derivatives of chemical potentials, followed by numerical integration along different compositions. Here we generalize the S0 method from two-component mixtures to neutral multi-component mixtures. We first extend the statistical mechanical formalism to high-dimensional compositional space, and then introduce a Gaussian process integration scheme combined with active learning to efficiently integrate chemical potentials and sample diverse compositions. We use this method to compute the mixing free energies of a molten metal alloy, and the solubilities of two paracetamol polymorphs in water-ethanol solvents. The extended S0 method provides a practical and scalable route for computing chemical potentials in neutral bulk multi-component mixtures from atomistic simulations.

Figures

Figures reproduced from arXiv: 2608.08357 by the authors.

**Figure 2.** FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** FIG. 3: A comparison of the chemical potential accuracy based on iteratively selecting new data points using CUR [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: FIG. 4: Free energy of mixing ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: FIG. 5: Excess chemical potential of paracetamol, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: FIG. 6: Solubility of paracetamol at 303.15 K and [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 52 canonical work pages

[1]

Sanz and C

E. Sanz and C. Vega, J. Chem. Phys.126, 014507 (2007)

work page 2007
[2]

A. S. Paluch, S. Jayaraman, J. K. Shah, and E. J. Mag- inn, J. Chem. Phys.133, 124504 (2010)

work page 2010
[3]

L ´ ısal, W

M. L ´ ısal, W. R. Smith, and J. Kolafa, J. Phys. Chem. B109, 12956 (2005)

work page 2005
[4]

Moucka, M

F. Moucka, M. L ´ ısal, J. Sˇkvor, J. Jirs´ ak, I. Nezbeda, and W. R. Smith, J. Phys. Chem. B115, 7849 (2011)

work page 2011
[5]

Perego, F

C. Perego, F. Giberti, and M. Parrinello, Eur. Phys. J.: Spec. Top.225, 1621 (2016)

work page 2016
[6]

I. S. Joung and T. E. Cheatham III, J. Phys. Chem. B 112, 9020 (2008)

work page 2008
[7]

L. Li, T. Totton, and D. Frenkel, J. Chem. Phys.146, 214110 (2017)

work page 2017
[8]

L. Li, T. Totton, and D. Frenkel, J. Chem. Phys.149, 054102 (2018)

work page 2018

Show all 53 references

[9]

H. A. Vinutha and D. Frenkel, J. Chem. Phys.154, 124502 (2021)

2021
[10]

M. P. Allen and D. J. Tildesley,Computer simulation in chemical physics, Vol. 397 (Springer Science & Business Media, 2012) pp. 108–111. 10

2012
[11]

Smit and D

B. Smit and D. Frenkel, Mol. Phys.68, 951 (1989)

1989
[12]

J. G. Kirkwood and F. P. Buff, J. Chem. Phys.19, 774 (1951)

1951
[13]

Dawass, P

N. Dawass, P. Kr¨ uger, S. K. Schnell, J.-M. Simon, and T. J. Vlugt, Fluid Phase Equilib.486, 21 (2019)

2019
[14]

Cortes-Huerto, K

R. Cortes-Huerto, K. Kremer, and R. Potestio, J. Chem. Phys.145, 141103 (2016)

2016
[15]

Cheng, J

B. Cheng, J. Chem. Phys.157, 121101 (2022)

2022
[16]

Reinhardt, P

A. Reinhardt, P. Y. Chew, and B. Cheng, J. Chem. Phys. 159, 184110 (2023)

2023
[17]

Herboth, D

R. Herboth, D. Dudariev, and A. P. Lyubartsev, Cryst. Growth Des.25, 7155 (2025)

2025
[18]

Cheng, S

B. Cheng, S. Hamel, and M. Bethkenhagen, Nat. Com- mun.14, 1104 (2023)

2023
[19]

X. Wang, S. Hamel, and B. Cheng, arXiv preprint arXiv:2603.28927 (2026)

2026
[20]

Schmid and B

R. Schmid and B. Cheng, J. Chem. Phys.158, 161101 (2023)

2023
[21]

Wang and B

X. Wang and B. Cheng, J. Chem. Phys.161, 034111 (2024)

2024
[22]

Ben-Naim,Molecular Theory of Solutions(Oxford University Press, Oxford, UK, 2006)

A. Ben-Naim,Molecular Theory of Solutions(Oxford University Press, Oxford, UK, 2006)

2006
[23]

J. W. Nichols, S. G. Moore, and D. R. Wheeler, Phys. Rev. E80, 051203 (2009)

2009
[24]

Busselez, J

R. Busselez, J. Chem. Phys.163, 134105 (2025)

2025
[25]

O’Connell, Mol

J. O’Connell, Mol. Phys.20, 27 (1971)

1971
[26]

H. B. Callen,Thermodynamics & an introduction to ther- mostatistics(John Wiley & Sons, 2006)

2006
[27]

Miryashkin, O

T. Miryashkin, O. Klimanova, V. Ladygin, and A. Shapeev, Phys. Rev. B108, 174103 (2023)

2023
[28]

V. L. Deringer, A. P. Bart´ ok, N. Bernstein, D. M. Wilkins, M. Ceriotti, and G. Cs´ anyi, Chem. Rev.121, 10073 (2021)

2021
[29]

Mones, N

L. Mones, N. Bernstein, and G. Cs´ anyi, J. Chem. Theory Comput.12, 5100 (2016)

2016
[30]

C. E. Rasmussen and C. K. I. Williams,Gaussian Pro- cesses for Machine Learning, Adaptive Computation and Machine Learning (MIT Press, Cambridge, MA, 2006)

2006
[31]

P. D. Sampson and P. Guttorp, Journal of the American Statistical Association87, 108 (1992)

1992
[32]

Snoek, K

J. Snoek, K. Swersky, R. S. Zemel, and R. P. Adams, inProceedings of the 31st International Conference on Machine Learning, Proceedings of Machine Learning Re- search, Vol. 32 (PMLR, Beijing, China, 2014) pp. 1674– 1682, arXiv:1402.0929 [stat.ML]

2014 arXiv
[33]

M. N. Gibbs,Bayesian Gaussian Processes for Regres- sion and Classification, Ph.D. thesis, University of Cam- bridge, Cambridge, UK (1997)

1997
[34]

C. J. Paciorek and M. J. Schervish, inAdvances in Neural Information Processing Systems 16, edited by S. Thrun, L. K. Saul, and B. Sch¨ olkopf (MIT Press, Cambridge, MA, 2004) pp. 273–280

2004
[35]

C. J. Paciorek and M. J. Schervish, Environmetrics17, 483 (2006)

2006
[36]

Remes, M

S. Remes, M. Heinonen, and S. Kaski, Advances in neu- ral information processing systems30(2017)

2017
[37]

M. W. Mahoney and P. Drineas, Proc. Natl. Acad. Sci. U. S. A.106, 697 (2009)

2009
[38]

D. R. Trinkle, Phys. Rev. Mater.9, 073801 (2025)

2025
[39]

A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolin- tineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimp- ton, Comput. Phys. Commun.271, 108171 (2022)

2022
[40]

Bonny, R

G. Bonny, R. C. Pasianot, N. Castin, and L. Malerba, Philos. Mag.89, 3531 (2009)

2009
[41]

Shinoda, M

W. Shinoda, M. Shiga, and M. Mikami, Phys. Rev. B 69, 134103 (2004)

2004
[42]

Bussi, D

G. Bussi, D. Donadio, and M. Parrinello, J. Chem. Phys. 126, 014101 (2007)

2007
[43]

Variankaval, A

N. Variankaval, A. S. Cote, and M. F. Doherty, AIChE J.54, 1682 (2008)

2008
[44]

Ruether and G

F. Ruether and G. Sadowski, J. Pharm. Sci.98, 4205 (2009)

2009
[45]

H.-H. Tung, E. L. Paul, M. Midler, and J. A. McCauley, Crystallization of organic compounds: an industrial per- spective(John Wiley & Sons, 2023)

2023
[46]

J. B. Klauda, R. M. Venable, J. A. Freites, J. W. O’Connor, D. J. Tobias, C. Mondragon-Ramirez, I. Vorobyov, A. D. J. MacKerell, and R. W. Pastor, J. Phys. Chem. B114, 7830 (2010)

2010
[47]

Nagai and S

T. Nagai and S. Prakongpan, Chem. Pharm. Bull.32, 340 (1984)

1984
[48]

Jouyban, O

A. Jouyban, O. Azarmir, S. Mirzaei, D. Hassan- zadeh, T. Ghafourian, J. William Eugene Acree, and A. Nokhodchi, Chem. Pharm. Bull.56, 602 (2008)

2008
[49]

G. P. Assis, R. H. L. Garcia, S. Derenzo, and A. Bernardo, J. Mol. Liq.323, 114617 (2021)

2021
[50]

M. A. Bellucci, G. Gobbo, T. K. Wijethunga, G. Ciccotti, and B. L. Trout, J. Chem. Phys.150, 094107 (2019)

2019
[51]

X. Ou, X. Li, H. Rong, L. Yu, and M. Lu, Chem. Com- mun.56, 9950 (2020)

2020
[52]

L. H. Thomas, C. Wales, L. Zhao, and C. C. Wilson, Cryst. Growth Des.11, 1450 (2011)

2011
[53]

Nishigaki, M

A. Nishigaki, M. Maruyama, S.-i. Tanaka, H. Y. Yoshikawa, M. Imanishi, M. Yoshimura, Y. Mori, and K. Takano, Crystals11, 1069 (2021)

2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Sanz and C

E. Sanz and C. Vega, J. Chem. Phys.126, 014507 (2007)

work page 2007

[2] [2]

A. S. Paluch, S. Jayaraman, J. K. Shah, and E. J. Mag- inn, J. Chem. Phys.133, 124504 (2010)

work page 2010

[3] [3]

L ´ ısal, W

M. L ´ ısal, W. R. Smith, and J. Kolafa, J. Phys. Chem. B109, 12956 (2005)

work page 2005

[4] [4]

Moucka, M

F. Moucka, M. L ´ ısal, J. Sˇkvor, J. Jirs´ ak, I. Nezbeda, and W. R. Smith, J. Phys. Chem. B115, 7849 (2011)

work page 2011

[5] [5]

Perego, F

C. Perego, F. Giberti, and M. Parrinello, Eur. Phys. J.: Spec. Top.225, 1621 (2016)

work page 2016

[6] [6]

I. S. Joung and T. E. Cheatham III, J. Phys. Chem. B 112, 9020 (2008)

work page 2008

[7] [7]

L. Li, T. Totton, and D. Frenkel, J. Chem. Phys.146, 214110 (2017)

work page 2017

[8] [8]

L. Li, T. Totton, and D. Frenkel, J. Chem. Phys.149, 054102 (2018)

work page 2018