pith. sign in

arxiv: 2606.25229 · v1 · pith:OKQ3OBQGnew · submitted 2026-06-23 · ❄️ cond-mat.stat-mech · physics.chem-ph

Accelerating Chemical Potential Calculations with Minimal Normalizing Flows

Philippe Baron , Athanassios Z. Panagiotopoulos This is my paper

Pith reviewed 2026-06-25 21:18 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech physics.chem-ph
keywords normalizing flowschemical potentialfree energy calculationmolecular simulationLennard-Jones fluidion solvationphase space overlapparticle insertion
0
0 comments X

The pith

Minimal normalizing flows with low-dimensional radial mappings accelerate chemical potential calculations by at least 10 times.

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

The paper shows that intentionally simple normalizing flows, limited to low-dimensional physically informed transformations such as radial scaling of particle positions, can map configurations between the Boltzmann distributions of non-inserted and inserted states. These minimal flows train in roughly one minute on a GPU thanks to a new training strategy, avoiding the convergence issues of more expressive networks. In Lennard-Jones particle systems the approach speeds chemical potential calculations by a factor of ten or more, while for ion solvation in water it raises effective sample size by factors of three for charging free energies and eight for force-field perturbation free energies. A reader cares because chemical potentials control solubility, phase behavior, and electrolyte properties, yet standard insertion methods suffer from vanishing phase-space overlap that makes them prohibitively expensive.

Core claim

A minimal normalizing flow applies a trainable but deliberately low-expressivity bijective mapping, built from radial (and optionally orientational) transformations, to improve overlap between the ensembles with and without an inserted particle or charge, thereby reducing the variance of free-energy estimates obtained from molecular simulations.

What carries the argument

A minimal normalizing flow (MNF): a trainable bijective mapping restricted to low-dimensional, physically informed transformations such as radial scaling that maps configurations between two Boltzmann distributions.

If this is right

  • Chemical potential calculations for Lennard-Jones fluids achieve at least a 10-fold reduction in computational cost.
  • Charging free-energy calculations for ions in water gain a 3-fold increase in effective sample size.
  • Free-energy differences arising from force-field perturbations gain an 8-fold increase in effective sample size.
  • Training completes in approximately one minute of GPU time, enabling on-the-fly use for new systems.
  • The same low-dimensional strategy supplies a practical route to physically informed mappings for other free-energy problems.

Where Pith is reading between the lines

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

  • The same radial-mapping idea could be tested on other alchemical transformations such as mutating one atom type into another.
  • Combining MNFs with existing enhanced-sampling methods might compound the efficiency gains for larger biomolecules.
  • If the radial transformation proves sufficient for many liquids, it suggests that domain knowledge can substitute for network depth in distribution-mapping tasks.
  • The method's low training cost opens the possibility of adapting the mapping continuously during a long simulation rather than training once upfront.

Load-bearing premise

Low-dimensional radial and orientational transformations are expressive enough to produce useful mappings between the relevant inserted and non-inserted Boltzmann distributions without overfitting or convergence failure.

What would settle it

Repeated tests on Lennard-Jones systems showing acceleration factors below 2x or training times exceeding a few minutes on standard GPU hardware would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2606.25229 by Athanassios Z. Panagiotopoulos, Philippe Baron.

Figure 1
Figure 1. Figure 1: Demonstration of 1D and 2D Minimal Normalizing Flows. (a) A one￾dimensional MNF, applied uniformly to every solvent particle a distance r from the inserted solute. The solvent particle is mapped to a new radial distance r ′ = ψ(r) = r + ∆r. (b) A two-dimensional MNF, where every oxygen at distance r away from the inserted ion is mapped to a distance r ′ = ψ(r) = r + ∆r while the water molecule is rotated a… view at source ↗
Figure 2
Figure 2. Figure 2: Mapped Work Distributions for the Chemical Potential of a Pure LJ Fluid. Both panels show the forward (blue) and reverse (orange) work distributions under the identity mapping, the ground truth chemical potential value (βµ, vertical dotted line), and an inset magnifying the overlap region (βW = βµ±5) with overlap shaded in yellow. (a) Forward (blue line) and Reverse (red line) mapped work distributions aft… view at source ↗
Figure 3
Figure 3. Figure 3: Acceleration of the Chemical Potential Calculation for a Binary LJ System. (a) Targeted free energy estimates (βµˆ) as a function of the number of independent samples collected from each state (Nind) for the identity mapping (grey), and mappings trained with the KLD (red) and BD-HD (purple) training approaches. (b) Standard error in the targeted free energy estimates (βσ(ˆµ)) as a function of the number of… view at source ↗
Figure 4
Figure 4. Figure 4: Acceleration of the Charging Free Energy Calculation for Na+ . (a) Charging free energy (∆Gcoul) versus window size (∆ϕ) for the identity mapping (blue) and the targeted estimate (orange) trained using the BD-HD procedure. (b) Effective sample size increase (˜n) versus window size (∆ϕ) for the BD-HD mapping compared to the identity mapping. The dashed line represents a power law fit to the data to guide th… view at source ↗
Figure 5
Figure 5. Figure 5: Mapped Work Distributions for the Free Energy Change of Adjusting the Na+ FF. Forward (light blue) and reverse (light orange) work distributions under the identity mapping are shown. The ground truth chemical potential value (βµ) is given by the vertical dotted line, and an inset magnifying the overlap region (βW = βµ ± 5) is displayed with overlap shaded in yellow. The forward and reverse mapped work dist… view at source ↗
read the original abstract

Chemical potentials are among the most important properties that can be obtained from a molecular simulation since they define many technologically relevant collective properties. The chemical potential of a species in solution is obtained by computing the free energy change of adding that species into a bulk system, a calculation typically very expensive for systems such as electrolytes, due to the lack of phase space overlap between "not-inserted" and "inserted" states. Recently, normalizing flows have been introduced as a way to accelerate free energy computations by learning a bijective function, constructed to be as expressive as possible, that maps the configuration space of one Boltzmann distribution onto another. This expressivity makes them difficult to train, limiting their ability to be generated "on-the-fly" for any new system, and in practice these mappings have shown only modest sampling improvements for liquids. We address these issues by introducing a "minimal" normalizing flow (MNF). This is a trainable bijective mapping that is intentionally limited in expressivity, and instead applies low-dimensional, physically informed transformations. Useful MNFs can be trained in 1 minute of GPU time due to their simplicity and our introduction of a novel training strategy. We show how calculations of chemical potentials of Lennard-Jones particle systems can be accelerated by at least 10 times with a simple radial mapping. We also apply a radial and orientational mapping to ion solvation in water, showing that MNFs can increase the effective sample size by 3 times for charging free energy calculations and 8 times for calculating free energy changes due to force field perturbations. This provides the foundation for the development of physically-informed mappings that can accelerate complex free energy calculations while retaining low training costs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces Minimal Normalizing Flows (MNFs) as intentionally low-expressivity bijective mappings that apply low-dimensional physically informed transformations (radial scaling for Lennard-Jones particle insertions; radial plus orientational for ion solvation in water). It claims these mappings, trained via a novel strategy in ~1 min of GPU time, accelerate chemical-potential calculations by at least 10× for LJ systems and increase effective sample size by 3× (charging free energies) and 8× (force-field perturbations) for aqueous ions, while avoiding the training difficulties of more expressive normalizing flows.

Significance. If the reported speedups are shown to be unbiased and reproducible with standard error controls, the work would supply a practical, low-cost route to improving phase-space overlap in free-energy calculations for liquids and electrolytes. The emphasis on minimal, physics-informed transformations and rapid training distinguishes it from prior normalizing-flow applications that have shown only modest gains in condensed-phase systems.

major comments (3)
  1. [Abstract / Results] Abstract and Results: the central performance claims (≥10× acceleration for LJ; 3× and 8× ESS gains for water) are presented without error bars, explicit baseline definitions, or the precise functional form of the radial/orientational mappings, preventing verification that the Jacobian-corrected estimates remain unbiased relative to direct insertion or thermodynamic integration.
  2. [Methods] Methods: the novel training strategy is asserted to avoid overfitting, yet no diagnostics (train/validation loss curves, ESS saturation versus training-set size, or comparison to an independent reference free-energy calculation) are supplied to confirm that the learned map has converged to the true optimal transport rather than an incomplete approximation.
  3. [Results (ion solvation)] Water/electrolyte results: the assumption that a single radial scaling plus rigid-body rotation suffices to capture the dominant distribution shift is load-bearing for the 3×/8× claims, but many-body polarization, hydrogen-bond rearrangement, and long-range screening are not obviously spanned by these coordinates; residual mismatch would produce systematic bias in the estimated charging or perturbation free energies.
minor comments (2)
  1. [Methods] Notation for effective sample size and the precise definition of the MNF Jacobian should be stated explicitly in the main text rather than left to supplementary material.
  2. [Figures] Figure captions should include the number of independent runs and the precise definition of the baseline (e.g., standard Widom insertion) used for the reported speedups.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive report and the opportunity to respond. We address each major comment below, indicating revisions where the manuscript will be strengthened.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results: the central performance claims (≥10× acceleration for LJ; 3× and 8× ESS gains for water) are presented without error bars, explicit baseline definitions, or the precise functional form of the radial/orientational mappings, preventing verification that the Jacobian-corrected estimates remain unbiased relative to direct insertion or thermodynamic integration.

    Authors: We agree that error bars, explicit baseline definitions, and expanded descriptions of the mappings will improve verifiability. The radial and orientational transformations are defined in the Methods, but we will add explicit functional forms and equations. Error bars from independent replicate runs will be added to all performance metrics, and baselines (direct insertion, thermodynamic integration) will be stated clearly. The Jacobian correction is derived to preserve unbiasedness, and we will include a short verification against reference calculations. These changes will be incorporated in the revised manuscript. revision: yes

  2. Referee: [Methods] Methods: the novel training strategy is asserted to avoid overfitting, yet no diagnostics (train/validation loss curves, ESS saturation versus training-set size, or comparison to an independent reference free-energy calculation) are supplied to confirm that the learned map has converged to the true optimal transport rather than an incomplete approximation.

    Authors: The intentionally low expressivity of MNFs combined with the novel training strategy is designed to promote rapid, stable convergence without overfitting. We acknowledge that explicit diagnostics would strengthen this claim. In revision we will add training and validation loss curves, plots of ESS versus training-set size, and direct comparisons of the resulting free energies to independent reference calculations (thermodynamic integration) to demonstrate convergence to the optimal transport. revision: yes

  3. Referee: [Results (ion solvation)] Water/electrolyte results: the assumption that a single radial scaling plus rigid-body rotation suffices to capture the dominant distribution shift is load-bearing for the 3×/8× claims, but many-body polarization, hydrogen-bond rearrangement, and long-range screening are not obviously spanned by these coordinates; residual mismatch would produce systematic bias in the estimated charging or perturbation free energies.

    Authors: The chosen low-dimensional transformations are physics-informed approximations that target the dominant radial and orientational shifts; the reported ESS gains indicate they are effective for the quantities studied. We recognize that many-body effects are not explicitly spanned and could introduce residual bias in more demanding regimes. In the revision we will add an explicit limitations paragraph discussing these effects, the conditions under which the approximation remains accurate, and how more expressive mappings could be substituted when needed. The core numerical claims will remain unchanged as they are supported by the observed improvements. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical speedups are measured outcomes, not tautological reductions

full rationale

The paper introduces MNFs as intentionally restricted, physically motivated low-dimensional maps (radial for LJ; radial+orientational for ions) and reports measured performance gains (10x acceleration, 3x/8x effective sample size increases) from direct simulation comparisons. These quantities are obtained by applying the trained bijections to compute free-energy differences and are not defined by or equivalent to the training loss or fitted parameters themselves. No equations reduce a claimed result to its own inputs by construction, no self-citation chain bears the central claim, and no uniqueness theorem is invoked. The work is an empirical demonstration of a new sampling technique whose validity rests on numerical benchmarks rather than definitional equivalence.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract-only review; no explicit free parameters, mathematical axioms, or invented physical entities are stated. The MNF itself is the primary new construct introduced.

invented entities (1)
  • Minimal Normalizing Flow (MNF) no independent evidence
    purpose: Trainable bijective mapping with deliberately limited expressivity that applies low-dimensional physically informed transformations to map between Boltzmann distributions for free-energy calculations
    Introduced in the abstract as the core new method to solve training difficulties of standard normalizing flows.

pith-pipeline@v0.9.1-grok · 5840 in / 1295 out tokens · 30315 ms · 2026-06-25T21:18:46.135334+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

178 extracted references · 3 canonical work pages

  1. [1]

    James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Yash Katariya and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

  2. [2]

    Berendsen, H. J. C. and Grigera, J. R. and Straatsma, T. P. , title =. J. Phys. Chem. , volume =. doi:10.1021/j100308a038 , year =

    work page doi:10.1021/j100308a038

  3. [3]

    , TITLE =

    Bhattacharyya, A. , TITLE =. Bull. Calcutta Math. Soc. , FJOURNAL =. 1943 , PAGES =

    1943

  4. [4]

    and Smit, B

    Frenkel, D. and Smit, B. , title =. 2002 , isbn =

    2002

  5. [5]

    McQuarrie, D. A. , title =. 2000 , isbn =

    2000

  6. [6]

    Chen, W and Ferguson, AL , title =. J. Comput. Chem. , volume =. doi:10.1002/jcc.25520 , year =

    work page doi:10.1002/jcc.25520

  7. [7]

    Shirts, MR and Bair, E and Hooker, G and Pande, VS , title =. Phys. Rev. Lett. , volume =. doi:10.1103/PhysRevLett.91.140601 , year =

    work page doi:10.1103/physrevlett.91.140601

  8. [8]

    Baron, P. B. and Panagiotopoulos, A. Z. , title =. J. Chem. Phys. , volume =. 2025 , doi =

    2025

  9. [9]

    Crooks, G. E. , title =. Phys. Rev. E , volume =. 2000 , doi =

    2000

  10. [10]

    and Hummer, G

    Dellago, C. and Hummer, G. , title =. Entropy , volume =. 2014 , doi =

    2014

  11. [11]

    , title =

    Jarzynski, C. , title =. Phys. Rev. Lett. , volume =. 1997 , doi =

    1997

  12. [12]

    and Nalisnick, E

    Papamakarios, G. and Nalisnick, E. and Rezende, D. J. and Mohamed, S. and Lakshminarayanan, B. , title =. J. Mach. Learn. Res. , volume =

  13. [13]

    Pham, T. T. and Shirts, M. R. , title =. J. Chem. Phys. , volume =. 2012 , doi =

    2012

  14. [14]

    and Kuznets-Speck, B

    Zhong, A. and Kuznets-Speck, B. and DeWeese, M. R. , title =. Phys. Rev. E , volume =. 2024 , doi =

    2024

  15. [15]

    and Kim, S

    Aykol, M. and Kim, S. and Hegde, V. I. and Snydacker, D. and Lu, Z. and Hao, S. Q. and Kirklin, S. and Morgan, D. and Wolverton, C. , title =. Nat. Commun. , volume =. 2016 , doi =

    2016

  16. [16]

    and Bekasov, A

    Durkan, C. and Bekasov, A. and Murray, I. and Papamakarios, G. , title =. Advances in Neural Information Processing Systems , editor =. 2019 , volume =

    2019

  17. [17]

    and Serra, E

    Ghidini, A. and Serra, E. and Cavalli, A. , title =. Acc. Chem. Res. , volume =. 2025 , doi =

    2025

  18. [18]

    Panagiotopoulos, A. Z. , title =. J. Phys. Condens. Matter , volume =. 2000 , doi =

    2000

  19. [19]

    Sadybekov, A. V. and Katritch, V. , title =. Nature , volume =. 2023 , doi =

    2023

  20. [20]

    and Sanz, E

    Vega, C. and Sanz, E. and Abascal, J. L. F. and Noya, E. G. , title =. J. Phys. Condens. Matter , volume =. 2008 , doi =

    2008

  21. [21]

    and Cai, W

    Ahmad, R. and Cai, W. , title =. Modell. Simul. Mater. Sci. Eng. , volume =. 2022 , doi =

    2022

  22. [22]

    Klein, L. and Kr. Equivariant Flow Matching , booktitle =. 2023 , volume =

    2023

  23. [23]

    and Papamakarios, G

    Wirnsberger, P. and Papamakarios, G. and Ibarz, B. and Racanière, S. and Ballard, A. J. and Pritzel, A. and Blundell, C. , title =. Mach. Learn.: Sci. Technol. , volume =. 2022 , doi =

    2022

  24. [24]

    Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning , journal =

    No. Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning , journal =. 2019 , doi =

    2019

  25. [25]

    and Frob

    Schebek, M. and Frob. Estimating solvation free energies with. 2512.18147 , eprintclass =

    arXiv

  26. [26]

    Midgley, L. I. and Stimper, V. and Simm, G. N. C. and Sch. Flow Annealed Importance Sampling Bootstrap , booktitle =. 2023 , eprinttype =. 2208.01893 , url =

    arXiv 2023

  27. [27]

    Felardos, L. and H. Designing losses for data-free training of normalizing flows on. 2301.05475 , eprintclass =

    arXiv

  28. [28]

    Schebek, M. and No. Scalable. 2509.25486 , eprintclass =

    arXiv

  29. [29]

    and Liu, Y

    Olehnovics, E. and Liu, Y. M. and Mehio, N. and Sheikh, A. Y. and Shirts, M. R. and Salvalaglio, M. , title =. J. Chem. Theory Comput. , volume =. 2024 , doi =

    2024

  30. [30]

    and Liu, Y

    Olehnovics, E. and Liu, Y. M. and Mehio, N. and Sheikh, A. Y. and Shirts, M. R. and Salvalaglio, M. , title =. J. Chem. Theory Comput. , volume =. 2025 , doi =

    2025

  31. [31]

    and Shirts, M

    Paliwal, H. and Shirts, M. R. , title =. J. Chem. Phys. , volume =. 2013 , doi =

    2013

  32. [32]

    and Shirts, M

    Paliwal, H. and Shirts, M. R. , title =. J. Chem. Theory Comput. , volume =. 2025 , doi =

    2025

  33. [33]

    and Carloni, P

    Rizzi, A. and Carloni, P. and Parrinello, M. , title =. J. Phys. Chem. Lett. , volume =. 2021 , doi =

    2021

  34. [34]

    and He, J

    Schebek, M. and He, J. J. and Hoffmann, E. and Du, Y. Q. and Noé, F. and Rogal, J. , title =. J. Chem. Phys. , volume =. 2026 , doi =

    2026

  35. [35]

    Schieber, N. P. and Shirts, M. R. , title =. J. Chem. Phys. , volume =. 2019 , doi =

    2019

  36. [36]

    Tan, C. B. and Bose, A. J. and Lin, C. and Klein, L. and Bronstein, M. M. and Tong, A. , title =. Proceedings of the 42nd International Conference on Machine Learning , editor =. 2025 , volume =

    2025

  37. [37]

    Tan, T. B. and Schultz, A. J. and Kofke, D. A. , title =. J. Chem. Phys. , volume =. 2010 , doi =

    2010

  38. [38]

    Willow, S. Y. and Kang, L. L. and Minh, D. D. L. , title =. J. Chem. Phys. , volume =. 2023 , doi =

    2023

  39. [39]

    and Falkner, S

    Coretti, A. and Falkner, S. and Geissler, P. L. and Dellago, C. , title =. J. Chem. Phys. , volume =. 2025 , doi =

    2025

  40. [40]

    Hahn, A. M. and Then, H. , title =. Phys. Rev. E , volume =. 2009 , doi =

    2009

  41. [41]

    , title =

    Jarzynski, C. , title =. Phys. Rev. E , volume =. 2002 , doi =

    2002

  42. [42]

    and Invernizzi, M

    Schebek, M. and Invernizzi, M. and Noé, F. and Rogal, J. , title =. Mach. Learn.: Sci. Technol. , volume =. 2024 , doi =

    2024

  43. [43]

    and Ballard, A

    Wirnsberger, P. and Ballard, A. J. and Papamakarios, G. and Abercrombie, S. and Racanière, S. and Pritzel, A. and Rezende, D. J. and Blundell, C. , title =. J. Chem. Phys. , volume =. 2020 , doi =

    2020

  44. [44]

    Chase, M. W., Jr. , title =. 1998 , address =

    1998

  45. [45]

    and Dicu-Gohoreanu, C

    Baran, L. and Dicu-Gohoreanu, C. A. and MacDowell, L. G. , title =. J. Chem. Phys. , volume =. 2025 , doi =

    2025

  46. [46]

    and Rahman, A

    Parrinello, M. and Rahman, A. , title =. J. Appl. Phys. , volume =. 1981 , doi =

    1981

  47. [47]

    Abascal, J. L. F. and Vega, C. , title =. J. Chem. Phys. , volume =. 2005 , doi =

    2005

  48. [48]

    Addula, R. K. R. and Veesam, S. K. and Punnathanam, S. N. , title =. Mol. Simul. , volume =. 2021 , doi =

    2021

  49. [49]

    and Marin-Rimoldi, E

    Agbodekhe, B. and Marin-Rimoldi, E. and Zhang, Y. and Dowling, A. W. and Maginn, E. J. , title =. J. Chem. Eng. Data , year =

  50. [50]

    Aktulga, H. M. and Kaymak, M. C. and Rahnamoun, A. and O'Hearn, K. A. and Merz, K. M. Jr and van Duin, A. C. T. , title =. J. Chem. Theory Comput. , volume =. 2022 , doi =

    2022

  51. [51]

    and Wang, P

    Anderko, A. and Wang, P. M. and Rafal, M. , title =. Fluid Phase Equilib. , volume =. 2002 , doi =

    2002

  52. [52]

    Aragones, J. L. and Noya, E. G. and Valeriani, C. and Vega, C. , title =. J. Chem. Phys. , volume =. 2013 , doi =

    2013

  53. [53]

    Aragones, J. L. and Valeriani, C. and Vega, C. , title =. J. Chem. Phys. , volume =. 2012 , doi =

    2012

  54. [54]

    Arora, J. S. and Elwakeil, O. A. and Chahande, A. I. and Hsieh, C. C. , title =. Struct. Optim. , volume =

  55. [55]

    and Musaelian, A

    Batzner, S. and Musaelian, A. and Sun, L. X. and Geiger, M. and Mailoa, J. P. and Kornbluth, M. and Molinari, N. and Smidt, T. E. and Kozinsky, B. , title =. Nat. Commun. , volume =. 2022 , doi =

    2022

  56. [56]

    Befort, B. J. and DeFever, R. S. and Tow, G. M. and Dowling, A. W. and Maginn, E. J. , title =. J. Chem. Inf. Model. , volume =. 2021 , doi =

    2021

  57. [57]

    Benavides, A. L. and Aragones, J. L. and Vega, C. , title =. J. Chem. Phys. , volume =. 2016 , doi =

    2016

  58. [58]

    Benavides, A. L. and Portillo, M. A. and Chamorro, V. C. and Espinosa, J. R. and Abascal, J. L. F. and Vega, C. , title =. J. Chem. Phys. , volume =. 2017 , doi =

    2017

  59. [59]

    Bennett, C. H. , title =. J. Comput. Phys. , volume =. 1976 , doi =

    1976

  60. [60]

    Berkowitz, M. L. , title =. J. Phys. Chem. B , volume =. 2021 , doi =

    2021

  61. [61]

    and Conde, M

    Bianco, V. and Conde, M. M. and Lamas, C. P. and Noya, E. G. and Sanz, E. , title =. J. Chem. Phys. , volume =. 2022 , doi =

    2022

  62. [62]

    and Conde, M

    Blazquez, S. and Conde, M. M. and Vega, C. , title =. J. Chem. Phys. , volume =. 2023 , doi =

    2023

  63. [63]

    and Kumar, R

    Carrillo-Bohórquez, O. and Kumar, R. and Kuroda, D. G. , title =. J. Phys. Chem. C , volume =. 2024 , doi =

    2024

  64. [64]

    and Zhu, X

    Caruso, A. and Zhu, X. Y. and Fulton, J. L. and Paesani, F. , title =. J. Phys. Chem. B , volume =. 2022 , doi =

    2022

  65. [65]

    and Noya, E

    Cassinello, G. and Noya, E. G. and Sanz, E. and Lamas, C. P. , title =. Mol. Phys. , volume =. 2024 , doi =

    2024

  66. [66]

    and Mandalaparthy, V

    Chattopadhyay, A. and Mandalaparthy, V. and van der Vegt, N. F. A. , title =. J. Chem. Phys. , volume =. 2025 , doi =

    2025

  67. [67]

    and Ko, H

    Chen, M. and Ko, H. Y. and Remsing, R. C. and Andrade, M. F. C. and Santra, B. and Sun, Z. R. and Selloni, A. and Car, R. and Klein, M. L. and Perdew, J. P. and Wu, X. F. , title =. Proc. Natl. Acad. Sci. U. S. A. , volume =. 2017 , doi =

    2017

  68. [68]

    Chodera, J. D. and Swope, W. C. and Pitera, J. W. and Seok, C. and Dill, K. A. , title =. J. Chem. Theory Comput. , volume =. 2007 , doi =

    2007

  69. [69]

    Cramer, S. D. , title =. Ind. Eng. Chem. Process Des. Dev. , volume =

  70. [70]

    Crooks, G. E. , title =. Phys. Rev. Lett. , volume =. 2007 , doi =

    2007

  71. [71]

    Davies, C. W. , title =. J. Chem. Soc. , pages =

  72. [72]

    and Intan, N

    Dinpajooh, M. and Intan, N. N. and Duignan, T. T. and Biasin, E. and Fulton, J. L. and Kathmann, S. M. and Schenter, G. K. and Mundy, C. J. , title =. J. Chem. Phys. , volume =. 2024 , doi =

    2024

  73. [73]

    and Lisal, M

    Dockal, J. and Lisal, M. and Moucka, F. , title =. J. Chem. Theory Comput. , volume =. 2020 , doi =

    2020

  74. [74]

    and Lisal, M

    Dockal, J. and Lisal, M. and Moucka, F. , title =. J. Mol. Liq. , volume =. 2022 , doi =

    2022

  75. [75]

    and Geissler, P

    Dodin, A. and Geissler, P. L. , title =. J. Chem. Theory Comput. , volume =. 2023 , doi =

    2023

  76. [76]

    and Brotzakis, Z

    Dutta, R. and Brotzakis, Z. F. and Mira, A. , title =. J. Chem. Phys. , volume =. 2018 , doi =

    2018

  77. [77]

    Espinosa, J. R. and Young, J. M. and Jiang, H. and Gupta, D. and Vega, C. and Sanz, E. and Debenedetti, P. G. and Panagiotopoulos, A. Z. , title =. J. Chem. Phys. , volume =. 2016 , doi =

    2016

  78. [78]

    and Alvarez, J

    Fernández-Prini, R. and Alvarez, J. L. and Harvey, A. H. , title =. J. Phys. Chem. Ref. Data , volume =. 2003 , doi =

    2003

  79. [79]

    Gee, M. B. and Cox, N. R. and Jiao, Y. F. and Bentenitis, N. and Weerasinghe, S. and Smith, P. E. , title =. J. Chem. Theory Comput. , volume =. 2011 , doi =

    2011

  80. [80]

    and Polat, H

    Habibi, P. and Polat, H. M. and Blazquez, S. and Vega, C. and Dey, P. and Vlugt, T. J. H. and Moultos, O. A. , title =. J. Phys. Chem. Lett. , volume =. 2024 , doi =

    2024

Showing first 80 references.