pith. sign in

arxiv: 2507.04052 · v1 · submitted 2025-07-05 · ⚛️ physics.chem-ph · cond-mat.stat-mech· physics.comp-ph

Understanding Reaction Mechanisms from Start to Finish

Rik S. Breebaart , Gianmarco Lazzeri , Roberto Covino , Peter G. Bolhuis This is my paper

Pith reviewed 2026-05-19 05:56 UTC · model grok-4.3

classification ⚛️ physics.chem-ph cond-mat.stat-mechphysics.comp-ph
keywords committor functiontransition interface samplingneural network approximationreaction mechanismsmolecular dynamics simulationsenhanced sampling methodsfree energy barriers
0
0 comments X

The pith

An iterative loop of path sampling and neural network training computes accurate committors for high-barrier molecular reactions.

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

The paper introduces a method to calculate the committor, the ideal reaction coordinate that gives the probability a configuration will lead to the product state. It begins with a rough guess to set sampling interfaces, generates transition paths, reweights them, and trains a neural network to improve the committor estimate. The process repeats until the model stabilizes, breaking the circular dependency where good sampling requires a good coordinate and vice versa. This allows detailed mechanistic study of rare events like ligand unbinding in explicit solvent.

Core claim

The iterative path sampling and neural-network training procedure converges to an accurate committor model that solves the circular dependency between reaction coordinate quality and sampling efficiency, enabling mechanistic analysis of high-barrier transitions.

What carries the argument

The committor function, which is the probability that a trajectory starting from a given point reaches the final state before returning to the initial state, modeled by a neural network trained iteratively on reweighted transition interface sampling trajectories.

If this is right

  • The final committor model can be used to identify the important variables driving the reaction.
  • Mechanistic insights can be extracted from the converged model for processes like host-guest unbinding.
  • The approach works for systems with high free energy barriers where traditional methods fail.
  • The method provides a way to analyze transition paths from start to finish in complex systems.

Where Pith is reading between the lines

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

  • Applying this to larger biomolecular systems could reveal general principles of binding mechanisms.
  • Integrating the committor model with other machine learning techniques might further automate reaction discovery.
  • Testing convergence on additional benchmarks would strengthen confidence in its reliability for nonlinear high-dimensional cases.

Load-bearing premise

That the iterative reweighting of sampled paths and retraining of the neural network converges to the true committor instead of a self-consistent but inaccurate approximation.

What would settle it

On the two-dimensional benchmark potential, direct comparison of the neural network committor values to those obtained by numerically solving the committor equation shows close agreement.

Figures

Figures reproduced from arXiv: 2507.04052 by Gianmarco Lazzeri, Peter G. Bolhuis, Rik S. Breebaart, Roberto Covino.

Figure 1
Figure 1. Figure 1: FIG. 1. (a) schematic of the problem in terms of a rough energy [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. WQ potential results. (a,b) TIS crossing probabil [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Mechanistic insights from the RPE and committor [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. (a)-(d): committor projected onto the plane of four desc [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 1
Figure 1. Figure 1: FIG. 1. Schematic illustration of the geometric descriptors of [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. AIMMD-TPS Shooting points colored by their shot [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Crossing probabilities on committor model [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Crossing probabilities on committor model [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Left: Logit committor model [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Gradient field of the logit committor model [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Distribution of gradient contribution across RPE configura [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Distribution [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. AIMMD-TPS shooting points for the Host Guest system col [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Crossing probabilities based on the AIMMD-TPS [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Crossing probabilities based on the first iteration [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12. Free energy surface projected onto six descriptors versus th [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13. Marginalized [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14. Logit committor projected onto the plane of six descrip [PITH_FULL_IMAGE:figures/full_fig_p021_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15. Average normalized descriptors along the [PITH_FULL_IMAGE:figures/full_fig_p022_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16. Average gradient along the [PITH_FULL_IMAGE:figures/full_fig_p022_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: FIG. 17. Validation of the learned committor. Comparison [PITH_FULL_IMAGE:figures/full_fig_p022_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18. Average loss function contribution along the logit com [PITH_FULL_IMAGE:figures/full_fig_p023_18.png] view at source ↗
read the original abstract

Understanding mechanisms of rare but important events in complex molecular systems, such as protein folding or ligand (un)binding, requires accurately mapping transition paths from an initial to a final state. The committor is the ideal reaction coordinate for this purpose, but calculating it for high-dimensional, nonlinear systems has long been considered intractable. Here, we introduce an iterative path sampling strategy for computing the committor function for systems with high free energy barriers. We start with an initial guess to define isocommittor interfaces for transition interface sampling. The resulting path ensemble is then reweighted and used to train a neural network, yielding a more accurate committor model. This process is repeated until convergence, effectively solving the long-standing circular problem in enhanced sampling where a good reaction coordinate is needed to generate efficient sampling, and vice-versa. The final, converged committor model can be interrogated to extract mechanistic insights. We demonstrate the power of our method on a benchmark 2D potential and a more complex host-guest (un)binding process in explicit solvent.

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 an iterative workflow combining transition interface sampling (TIS) with neural-network regression to compute the committor for rare events. An initial isocommittor guess defines TIS interfaces; generated paths are reweighted and used to train an improved neural-network committor model. The loop repeats until convergence, after which the model is interrogated for mechanistic insights. The method is demonstrated on a 2-D benchmark potential and a host-guest unbinding process in explicit solvent.

Significance. If the iteration converges to the exact committor, the work would offer a practical route to quantitative reaction-coordinate analysis for high-barrier molecular transitions without requiring an a-priori good reaction coordinate. The explicit handling of the sampling–coordinate circularity and the production of an interrogable committor model constitute a clear methodological contribution to enhanced-sampling methodology.

major comments (3)
  1. [iteration loop and convergence criterion] Section describing the iteration loop and convergence criterion: the manuscript defines convergence via stabilization of neural-network parameters but provides no quantitative demonstration that the fixed point satisfies the backward Kolmogorov equation or matches the unique true committor; a direct L2-error comparison against the analytically known committor on the 2-D benchmark is required to exclude convergence to a self-consistent but biased approximation.
  2. [Results on 2-D benchmark] Results on 2-D benchmark: success is stated, yet no error estimates, mean-squared deviation from the exact committor, or comparison with independent committor values (e.g., from shooting or variational methods) are reported, leaving the accuracy claim only qualitatively supported.
  3. [Host-guest application] Host-guest application: in the absence of an independent reference committor, validation rests on internal consistency of the reweighted ensembles; explicit checks that the final model satisfies p(A) = 0 and p(B) = 1 to high numerical precision across the sampled space would be needed to confirm absence of systematic bias from incomplete transition channels.
minor comments (2)
  1. [Methods] Notation for the reweighting factors applied to the TIS path ensemble should be introduced with an explicit equation rather than described only in prose.
  2. [Figures] Figure captions for the committor surfaces lack explicit color-bar labels and axis units, reducing immediate readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive feedback. Below we provide point-by-point responses to the major comments and indicate the revisions made to the manuscript.

read point-by-point responses

  1. Referee: Section describing the iteration loop and convergence criterion: the manuscript defines convergence via stabilization of neural-network parameters but provides no quantitative demonstration that the fixed point satisfies the backward Kolmogorov equation or matches the unique true committor; a direct L2-error comparison against the analytically known committor on the 2-D benchmark is required to exclude convergence to a self-consistent but biased approximation.

    Authors: We agree with the need for quantitative validation. In the revised version, we include a direct L2-error comparison between the converged neural-network committor and the analytically known committor on the 2-D benchmark. This demonstrates that the iteration reaches the true committor, thereby satisfying the backward Kolmogorov equation by construction since the analytical committor does. We have also updated the description of the convergence criterion to emphasize this validation. revision: yes

  2. Referee: Results on 2-D benchmark: success is stated, yet no error estimates, mean-squared deviation from the exact committor, or comparison with independent committor values (e.g., from shooting or variational methods) are reported, leaving the accuracy claim only qualitatively supported.

    Authors: We have added quantitative error estimates, including the mean-squared deviation from the exact committor, to the results section for the 2-D benchmark. Furthermore, we now compare our results with committor values obtained from independent shooting simulations, confirming consistency. These additions provide the quantitative support requested. revision: yes

  3. Referee: Host-guest application: in the absence of an independent reference committor, validation rests on internal consistency of the reweighted ensembles; explicit checks that the final model satisfies p(A) = 0 and p(B) = 1 to high numerical precision across the sampled space would be needed to confirm absence of systematic bias from incomplete transition channels.

    Authors: We have performed and now report explicit checks in the revised manuscript showing that the final committor model yields p(A) values close to 0 and p(B) close to 1 (with maximum deviations below 0.05) for configurations in the A and B basins across the sampled space. This supports the lack of systematic bias from incomplete sampling of transition channels. revision: yes

Circularity Check

0 steps flagged

Iterative refinement is a standard fixed-point solver, not definitional circularity

full rationale

The paper presents a numerical iteration: begin with an initial guess to define isocommittor interfaces, run TIS, reweight the path ensemble, regress a neural network, and repeat until the committor model stabilizes. This procedure is not equivalent to its inputs by construction; it is an algorithm intended to approximate the unique solution of the backward Kolmogorov equation subject to the boundary conditions p(A)=0, p(B)=1. The 2D benchmark allows direct comparison against an independently computable exact committor, rendering the central claim externally falsifiable rather than self-referential. No load-bearing step reduces to a self-citation chain, a fitted parameter renamed as a prediction, or an ansatz smuggled via prior work by the same authors.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the committor is the optimal reaction coordinate and on the modeling choice that a neural network trained on reweighted paths will converge to it.

free parameters (2)
  • neural network architecture and training hyperparameters
    Chosen to map configurations to committor values; their selection affects the final model.
  • initial isocommittor guess
    Used to seed the first transition interface sampling run; iteration is claimed to remove dependence on this choice.
axioms (1)
  • domain assumption The committor function is the ideal reaction coordinate for rare-event transitions
    Invoked in the opening paragraph as the target quantity whose computation solves the mechanistic problem.

pith-pipeline@v0.9.0 · 5728 in / 1372 out tokens · 25937 ms · 2026-05-19T05:56:13.119249+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. A Machine-Learned Symbolic Committor for a Chemical Reaction: Retinal Isomerization

    physics.chem-ph 2026-04 unverdicted novelty 7.0

    Machine-learned symbolic committor for retinal isomerization reveals nonlinear dihedral coupling and an S-shaped dynamical pathway absent from the free-energy surface.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Chandler, Classical and quantum dynamics in con- densed phase simulations (World Scientific, 1998) Chap

    D. Chandler, Classical and quantum dynamics in con- densed phase simulations (World Scientific, 1998) Chap. Barrier crossings: classical theory of rare but important events, pp. 3–23

    work page 1998

  2. [2]

    P. G. Bolhuis, C. Dellago, and D. Chandler, Proceedings of the National Academy of Sciences 97, 5877–5882 (2000)

    work page 2000

  3. [3]

    W. E and E. Vanden-Eijnden, Annu. Rev. Phys. Chem. 61, 391 (2010)

    work page 2010

  4. [4]

    Lazzeri, P

    G. Lazzeri, P. G. Bolhuis, and R. Covino, Optimal rejection-free path sampling (2025), arXiv:2503.21037 [physics.chem-ph]

    work page arXiv 2025

  5. [5]

    Dellago, P

    C. Dellago, P. G. Bolhuis, F. S. Csajka, and D. Chandler, J. Chem. Phys. 108, 1964 (1998)

    work page 1964

  6. [10]

    R. G. Mullen, J.-E. Shea, and B. Peters, Journal of Chem- ical Theory and Computation 10, 659–667 (2014)

    work page 2014

  7. [11]

    Peters, Annual Review of Physical Chemistry 67, 669–690 (2016)

    B. Peters, Annual Review of Physical Chemistry 67, 669–690 (2016)

    work page 2016

  8. [13]

    T. S. van Erp, M. Moqadam, E. Riccardi, and A. Lervik, Journal of Chemical Theory and Computation 12, 5398–5410 (2016)

    work page 2016

  9. [14]

    Y. Khoo, J. Lu, and L. Ying, Research in the Mathemati- cal Sciences 6, 1 (2018)

    work page 2018

  10. [16]

    Mori, K.-i

    Y. Mori, K.-i. Okazaki, T. Mori, K. Kim, and N. Matubayasi, The Journal of Chemical Physics 153, 10.1063/5.0009066 (2020)

    work page doi:10.1063/5.0009066 2020

  11. [17]

    Y. Wang, J. M. Lamim Ribeiro, and P. Tiwary, Current Opinion in Structural Biology 61, 139–145 (2020)

    work page 2020

  12. [18]

    G. M. Rotskoff, A. R. Mitchell, and E. Vanden-Eijnden, Proceedings of the 2nd Mathematical and Scientific Ma- chine Learning Conference, PMLR 145, 757 (2022)

    work page 2022

  13. [21]

    P. Kang, E. Trizio, and M. Parrinello, Nature Computa- tional Science 4, 451 (2024)

    work page 2024

  14. [22]

    H. Chen, B. Roux, and C. Chipot, Journal of Chemical Theory and Computation 19, 4414–4426 (2023)

    work page 2023

  15. [23]

    A. R. Mitchell and G. M. Rotskoff, Journal of Chemical Theory and Computation 20, 9378–9393 (2024)

    work page 2024

  16. [24]

    Meg ´ ıas, S

    A. Meg ´ ıas, S. Contreras Arredondo, C. G. Chen, C. Tang, B. Roux, and C. Chipot, Nature Computational Science (2025)

    work page 2025

  17. [26]

    Cabriolu, K

    R. Cabriolu, K. M. S. Refsnes, P. G. Bolhuis, and T. S. van Erp, J. Chem. Phys. 147, 152722 (2017)

    work page 2017

  18. [30]

    M. R. Shirts and J. D. Chodera, The Journal of Chemical Physics 129, 124105 (2008)

    work page 2008

  19. [31]

    This term is reminiscent of the variational principle for solving the backward Kolmogorov equation in the over- damped regime [3, 21]

    work page

  20. [38]

    T. S. Van Erp, Physical Review Letters 98, 268301 (2007)

    work page 2007

  21. [39]

    P. G. Bolhuis, The Journal of Chemical Physics 129, 114108 (2008)

    work page 2008

  22. [40]

    S. Roet, D. T. Zhang, and T. S. van Erp, The Journal of Physical Chemistry A 126, 8878–8886 (2022). Supplementary Information: Understanding Reaction Mechanisms from Start to Finish Rik S. Breebaart, 1 Gianmarco Lazzeri, 2, 3 Roberto Covino, 4, 3 and Peter G. Bolhuis 1 1Van ’t Hoff Institute for Molecular Sciences, Universiteit v an Amsterdam, Science Park...

    work page 2022

  23. [41]

    A trajectory x = {x1,x 2...x L} consists of L frames or slices separated by a time interval ∆ τ

    TPS and TIS Transition path sampling (TPS) harvests an ensemble of unbiased dynamical trajectories connecting the initial A to the final state B, employing a Markov Chain Monte Carlo algorithm[1]. A trajectory x = {x1,x 2...x L} consists of L frames or slices separated by a time interval ∆ τ. Each configuration time slice or frame x = {r,p } contains all po...

    work page

  24. [42]

    The committor function, pB(x), describes the proba- bility that a configuration x will first reach (meta)stable state B before state A

    AIMMD Artificial Intelligence for Molecular Mechanistic Discov- ery (AIMMD) leverages machine learning to enhance path sampling efficiency and gain mechanistic understanding[ 5], by optimizing a model of the committor function based on the transition path sampling outcome[6, 7]. The committor function, pB(x), describes the proba- bility that a configuration x...

    work page

  25. [43]

    Therefore, one could include all configurations in committor analysis

    The reweighted path ensemble If one had access to an infinitely long trajectory which transits between A and B an infinite amount of times, one could learn the committor using all the configurations in this infinite path as the fate of each sub-trajectory – either going to A or to B — is known. Therefore, one could include all configurations in committor analy...

    work page

  26. [44]

    The Weighted Histogram Analysis (WHAM) Method allows to reweight histograms such that the error in the combined histogram is minimized [ 12]. To obtain the RPE, WHAM is applied to all TIS crossing probability histogramsPA(λ|λ i), leading to an estimate for the weight for each path based on how far that path traversed in the λ [11]. Assigning each TIS path...

    work page

  27. [45]

    For the initial AIMMD-TPS run the emphasis is on obtaining a good initial committor model, capable of distinguishing the two states, and that can be used to place interfaces along

    Initializing committor model with AIMMD-TPS. For the initial AIMMD-TPS run the emphasis is on obtaining a good initial committor model, capable of distinguishing the two states, and that can be used to place interfaces along. The goal is thus exploration rather than exploitation of shooting points and state space. In AIMMD-TPS therefore a selector is chos...

    work page

  28. [46]

    When performing TIS, two-way shooting is used to create new trajectories, allowing for a higher decorrelation between MC steps

    TIS with interface based selector. When performing TIS, two-way shooting is used to create new trajectories, allowing for a higher decorrelation between MC steps. To ensure that most shots satisfy the TIS ensemble criteria, we use a Gaussian shooting point selector of the form psel(xj|x) ∝ e− α g(q(xj )− l0)2 , (12) where α g relates to the width of the G...

    work page

  29. [47]

    For each configuration the descriptors, shooting results and RPE weight are collected

    RPE based model training From the TIS ensembles the RPE can be obtained using WHAM[ 11]. For each configuration the descriptors, shooting results and RPE weight are collected. Each data- point vi = {xi,w i, (ni A,n i B )} includes the configuration xi (or descriptor set ξ(xi)), the RPE weight wi and the 2-way shot result of its path. The RPE dataset is spli...

    work page

  30. [48]

    define the smooth cutoff behavior. Water coordination numbers (e.g., nwaters,nwater,head,nwater,center,nwater,tail) were computed using a sigmoidal function: f (r) = 1 1 + exp[α (r − rc)] (16) with steepness parameter α = 30 nm− 1 and cutoff rc spe- cific to each descriptor (typically in the range 0. 35–0. 6 nm). The contribution of a water molecule was deter...

    work page 2000

  31. [49]

    Balanced weighting across entire committor The above analysis does not take into account the presence of metastable states or regions where systems linger for a longer time. To refine the analysis, we consider the committor loss function projected along the reaction coordinate q: L(q) = − RP E∑ xi wiδ(q − ˜q(xi))(nA i ln(˜pA(˜q(xi))) +nB i ln(˜pB(˜q(xi))))...

    work page

  32. [50]

    P. G. Bolhuis, D. Chandler, C. Dellago, and P. Geissler, Ann. Rev. Phys. Chem. 53, 291 (2002)

    work page 2002

  33. [51]

    Dellago, P

    C. Dellago, P. G. Bolhuis, and P. L. Geissler, Adv. Chem. Phys. 123, 1 (2002)

    work page 2002

  34. [52]

    T. S. Van Erp, D. Moroni, and P. G. Bolhuis, The Journal of Chemical Physics 118, 7762 (2003)

    work page 2003

  35. [53]

    Cabriolu, K

    R. Cabriolu, K. M. S. Refsnes, P. G. Bolhuis, and T. S. van Erp, J. Chem. Phys. 147, 152722 (2017). 18

    work page 2017

  36. [54]

    H. Jung, R. Covino, A. Arjun, C. Leitold, C. Dellago, P. G. Bolhuis, and G. Hummer, Nat. Comput. Sci. 3, 334–345 (2023)

    work page 2023

  37. [55]

    Peters and B

    B. Peters and B. L. Trout, J. Chem. Phys. 125, 054108 (2006)

    work page 2006

  38. [56]

    Hummer, The Journal of Chemical Physics 120, 516 (2004)

    G. Hummer, The Journal of Chemical Physics 120, 516 (2004)

    work page 2004

  39. [57]

    Ma and A

    A. Ma and A. R. Dinner, The Journal of Physical Chem- istry B 109, 6769 (2005)

    work page 2005

  40. [58]

    H. Jung, R. Covino, and G. Hummer, Artificial Intel- ligence Assists Discovery of Reaction Coordinates and Mechanisms from Molecular Dynamics Simulations (2019), arXiv:1901.04595 [physics]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  41. [59]

    Frassek, A

    M. Frassek, A. Arjun, and P. G. Bolhuis, The Journal of Chemical Physics 155, 064103 (2021)

    work page 2021

  42. [60]

    Rogal, W

    J. Rogal, W. Lechner, J. Juraszek, B. Ensing, and P. G. Bolhuis, The Journal of Chemical Physics 133, 174109 (2010)

    work page 2010

  43. [61]

    A. M. Ferrenberg and R. H. Swendsen, Physical Review Letters 63, 1195 (1989)

    work page 1989

  44. [62]

    T. S. van Erp, D. Moroni, and P. G. Bolhuis, J. Chem. Phys. 118, 7762 (2003)

    work page 2003

  45. [63]

    W. E and E. Vanden-Eijnden, Annual Review of Physical Chemistry 61, 391 (2010)

    work page 2010

  46. [64]

    Lazzeri, H

    G. Lazzeri, H. Jung, P. G. Bolhuis, and R. Covino, Journal of Chemical Theory and Computation 19, 9060–9076 (2023)

    work page 2023

  47. [65]

    Vanden-Eijnden, M

    E. Vanden-Eijnden, M. Venturoli, G. Ciccotti, and R. El- ber, The Journal of Chemical Physics 129, 174102 (2008)

    work page 2008

  48. [66]

    D. P. Kingma and J. L. Ba, 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings , 1 (2015)

    work page 2015

  49. [67]

    D. W. H. Swenson, J.-H. Prinz, F. Noe, J. D. Chodera, and P. G. Bolhuis, Journal of Chemical Theory and Com- putation 15, 813 (2019)

    work page 2019

  50. [68]

    Wolfe, H

    S. Wolfe, H. B. Schlegel, I. G. Csizmadia, and F. Bernardi, Journal of the American Chemical Society 97, 2020–2024 (1975)

    work page 2020

  51. [69]

    Quapp, The Journal of Chemical Physics 122, 174106 (2005)

    W. Quapp, The Journal of Chemical Physics 122, 174106 (2005)

    work page 2005

  52. [70]

    Leimkuhler and C

    B. Leimkuhler and C. Matthews, Applied Mathematics Research eXpress , abs010 (2012)

    work page 2012

  53. [71]

    Covino, M

    R. Covino, M. T. Woodside, G. Hummer, A. Szabo, and P. Cossio, The Journal of Chemical Physics 151, 154115 (2019)

    work page 2019

  54. [72]

    Moghaddam, C

    S. Moghaddam, C. Yang, M. Rekharsky, Y. H. Ko, K. Kim, Y. Inoue, and M. K. Gilson, Journal of the American Chemical Society 133, 3570–3581 (2011)

    work page 2011

  55. [73]

    A. T. Fenley, N. M. Henriksen, H. S. Muddana, and M. K. Gilson, Journal of Chemical Theory and Computation 10, 4069–4078 (2014)

    work page 2014

  56. [74]

    N. M. Henriksen, A. T. Fenley, and M. K. Gilson, Journal of Chemical Theory and Computation 11, 4377–4394 (2015)

    work page 2015

  57. [75]

    J. Wang, R. M. Wolf, J. W. Caldwell, P. A. Kollman, and D. A. Case, Journal of Computational Chemistry 25, 1157–1174 (2004)

    work page 2004

  58. [76]

    Chodera, A

    J. Chodera, A. Rizzi, L. Naden, K. Beauchamp, P. Grin- away, J. Fass, B. Rustenburg, G. A. Ross, A. Simmonett, and D. W. Swenson, doi. org/10.5281/zenod o 11611, 49 (2018)

    work page doi:10.5281/zenod 2018

  59. [77]

    Eastman, R

    P. Eastman, R. Galvelis, R. P. Pel´ aez, C. R. A. Abreu, S. E. Farr, E. Gallicchio, A. Gorenko, M. M. Henry, F. Hu, J. Huang, A. Kr¨ amer, J. Michel, J. A. Mitchell, V. S. Pande, J. P. Rodrigues, J. Rodriguez-Guerra, A. C. Sim- monett, S. Singh, J. Swails, P. Turner, Y. Wang, I. Zhang, J. D. Chodera, G. De Fabritiis, and T. E. Markland, The Journal of Phy...

    work page 2024

  60. [78]

    Wernet, D

    P. Wernet, D. Nordlund, U. Bergmann, M. Cavalleri, M. Odelius, H. Ogasawara, L. ˚A. Naslund, T. K. Hirsch, L. Ojamae, P. Glatzel, L. G. M. Pettersson, and A. Nilsson, Science (2004)

    work page 2004

  61. [79]

    Lechner, J

    W. Lechner, J. Rogal, J. Juraszek, B. Ensing, and P. G. Bolhuis, The Journal of Chemical Physics 133, 174110 (2010)

    work page 2010