Pith. sign in

REVIEW 4 major objections 5 minor 96 references

Boundary conditions can be dropped from the committor variational principle, making trial functions that violate them admissible.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 05:08 UTC pith:7SPXE2EE

load-bearing objection Genuinely new boundary-free variational principle, clean math, honest limitations; but the chignolin rate claim rests on an uncertified plateau and should be labeled preliminary. the 4 major comments →

arxiv 2608.02536 v1 pith:7SPXE2EE submitted 2026-08-03 physics.chem-ph

Committors and Reaction Rates from Trial Functions That Violate the Boundary Conditions

Magnus Petersen , Simon Lichtinger , Roberto Covino This is my paper
classification physics.chem-ph
keywords committortransition path theoryreactive fluxDirichlet energyvariational principlereaction coordinaterare eventsumbrella sampling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The committor—the probability that a trajectory reaches product before reactant—is the ideal reaction coordinate for rare transitions, but it is normally defined as the minimizer of a Dirichlet energy over functions that vanish on one state and equal one on the other. This paper proves that these boundary conditions can be replaced by a single scalar normalization, the fidelity, which is measurable from state-labelled equilibrium samples. Because the boundary conditions are gone, trial functions that cannot satisfy them—such as one-dimensional profiles along random projections—are admissible, and the optimal combination has a closed form. The resulting estimator consumes only configurations reweightable to equilibrium, two state definitions, and a diffusion constant, and the paper demonstrates it on peptides in full torsion space and on chignolin folding rates from umbrella sampling alone. The cost is roughly linear in sample count and dimension at fixed number of directions.

Core claim

The central claim is the identity ν_AB = min_u E[u]/F[u]^2, where E[u] is the Dirichlet energy and F[u], the fidelity, is the difference between the flux-weighted boundary averages of u. The identity follows from the flux-fidelity relation ⟨u,q⟩_D = ν_AB F[u], obtained by integrating the trial function against the conserved reactive current. The quotient is scale- and offset-free, the minimum is attained by u = λq + const, and no boundary condition is imposed. In a finite trial space the same principle yields an upper bound on the reactive flux that tightens monotonically as directions are added. Using basin-sample moments as empirical fidelities, the paper obtains high-dimensional committor

What carries the argument

The load-bearing object is the flux–fidelity identity: for any trial function u of finite Dirichlet energy, the inner product ⟨u,q⟩_D equals the reactive flux ν_AB times the fidelity F[u], defined as the flux-weighted average of u over the product boundary minus the flux-weighted average over the reactant boundary. This turns the Dirichlet error expansion into a ratio that needs no boundary conditions. The paper pairs it with a ridge-function ansatz: one-dimensional reaction-diffusion profiles q_j along Fisher-discriminant-broadened random projections, combined by the closed-form solve of a rank-one generalized eigenvalue problem, with regularization selected by a held-out variational score.

Load-bearing premise

The exact theorems hold for the true flux-weighted fidelity, but the practical estimator substitutes the difference of basin sample averages, which turns the bound into an approximation; the rate step additionally assumes a flat isocommittor-flux plateau that is explicitly violated in the chignolin application.

What would settle it

Run the estimator on a high-dimensional system with known committor (e.g., a harmonic barrier with many overlapping projections) and compare the predicted ν_AB against a long-trajectory forward-flux estimate. If the variational quotient E/F^2 ever falls below the true reactive flux, the identity is wrong; if the convergence in the number of directions stalls while the plateau diagnostic fails, the rate read-off needs revisiting.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Committors can be estimated from equilibrium or biased (umbrella-sampled, metadynamics) configurations alone, with no time-lagged pairs, shooting trajectories, or iterative sampling.
  • The variational bound survives without boundary conditions, so any trial function yields a certified upper bound on the reactive flux that can be tightened monotonically.
  • Rates can be recovered from biased data by reading the isocommittor flux plateau, as demonstrated for chignolin folding and unfolding from umbrella sampling alone.
  • The method scales linearly in sample count and dimension at fixed direction count, making full-torsion-space committors feasible for peptides of 52–350 dimensions.
  • The fidelity measurement provides a boundary diagnostic that can be evaluated for any variational committor method, not just this one.

Where Pith is reading between the lines

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

  • The same flux–fidelity identity should transfer to other reversible processes with two absorbing sets, such as nucleation, allele fixation, or climate transitions, where equilibrium-like samples exist but no clean boundaries can be imposed.
  • One testable extension is to position-dependent diffusion and non-reversible steady states: the paper's closed-form Gram matrix assumes constant diffusion, but the flux–fidelity identity itself does not.
  • The empirical-fidelity substitution (basin moments for flux-weighted averages) could be validated more generally by comparing against reactive-trajectory estimators on systems with many more committed events than villin's fifteen.
  • The plateau assumption in the rate read-off is the fragile step; a search for a flat isocommittor-flux region (or lack thereof) should become a standard diagnostic in any application.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper claims a new variational principle for committors and reaction rates in which the Dirichlet boundary conditions are replaced by a scalar normalization of a 'fidelity' functional F[u] that is measurable from state-labelled equilibrium samples. The exact identity E[q] = min_u E[u]/F[u]^2 (Eq. 5) is derived cleanly. The authors then propose a 'sliced committor' estimator built from 1D profiles along random projection directions, with a closed-form optimum (Eq. 13). They apply it to AIB9 and villin HP-35 committors in full torsion space and to chignolin folding/unfolding rates from umbrella sampling alone. The advertised practical contribution is that rates and committors can be obtained from equilibrium/reweighted samples without boundary-satisfying trial functions or dynamical trajectories. The exact mathematical identity is sound, but the estimator replaces the flux-weighted fidelity by a basin-moment difference (Eq. 10), and the chignolin rate read-off uses an isocommittor flux plateau that the paper's own diagnostics show to be absent.

Significance. If the numerical estimator were as reliable as the exact variational identity, this would be a significant advance: it removes the principal construction cost in variational committor methods and allows trial spaces that violate boundary conditions, including cheap 1D slices. The paper is commendably open with code, data, and extensive diagnostics; the supplementary material explicitly quantifies biases, reports negative gaps, and flags the chignolin plateau failure. These features raise the quality of the report. However, the significance of the practical claims is currently limited because the substitution in Eq. (10) is uncontrolled in high-dimensional molecular feature spaces, and the one rate application, chignolin, rests on a flux read-off that the authors themselves show to be non-plateauing. The exact identity and the algorithmic framework are publishable, but the molecular rate and error-bound claims are not yet supported at the level asserted.

major comments (4)
  1. [Main text, §'Fidelity from samples', Eq. (10)] The exact variational bound Eq. (5) and the certified cap Eq. (S6) are stated for the flux-weighted fidelity F[u]. The implementation replaces it by the equilibrium basin-moment difference \hat f_j = b_j - a_j. This is a different measure, not just statistical noise. The supplement ('Empirical fidelity on a molecular system') states that in the d=52–350 feature spaces 'almost every direction overlaps' and that the per-slice bias saturates near 0.15 on overlapping directions. The defense is weight suppression (Fig. S1c), demonstrated on two 2D systems and on AIB9, not on chignolin. For chignolin no reactive segments exist to evaluate the weighted deficit of Eq. (S9), so the central rate estimate has no check on this substitution. This is load-bearing and should be addressed by either supplying a reactive-flux fidelity estimate for chignolin or explicitly qualifying the rate as uncontrolle
  2. [Chignolin rate read-off, Eq. (S16)] The rate is read from the isocommittor flux plateau, but Supplement 'Chignolin: the plateau premise fails' reports that the profile decays monotonically, with flatness 0.71 (four times the largest of the other systems), band mass 27% versus the 60% expected from flux conservation, and 22% shifts between nested windows. The flatness-selected sub-band gives ehat = 1.76, and the paper states that the reported ehat should be read 'as a diagnostic ... and not as a calibrated error bar.' Despite this, the main text reports chignolin folding/unfolding rates of 2.2 and 0.20 µs^{-1} and claims agreement within a factor of 2.5. This is an internal inconsistency: the rate claim is not backed by the method's own certification.
  3. [AIB9 diagnosis in Supplement, 'Diagnosing the negative gap'] The supplementary analysis of AIB9 shows that the plateau read-off overestimates the flux by at least 9.7% (cap E/F^2 = 0.01819 versus plateau µν_AB = 0.01995), and the M=1024 rung gives r ≥ 1.29 and e ≥ 0.175. This demonstrates that the isocommittor plateau estimator of Eq. (S16) can be substantially biased high even when the pointwise committor RMSE is good. Since the same read-off is used for the chignolin rate, this is a direct concern for the paper's central numerical claim, independent of the chignolin plateau failure.
  4. [Main text, §'An error bound without the answer', Eq. (14)] Equation (14) is called an 'error bound without the answer,' but with the estimated \hat\nu_AB it is an estimate, not a bound; the supplement's Table S1 and the AIB9 analysis make this clear. The certified bound Eq. (S6) is only certified for F, not for \hat f. The paper oscillates between 'bound' and 'estimate' language. This is not merely a wording issue: it obscures the fact that the central error statement is uncontrolled in the chignolin application. Please either prove a bound for the empirical estimator under explicit assumptions or consistently present Eq. (14) as a diagnostic.
minor comments (5)
  1. [Notation throughout] The symbol F is used both for the fidelity functional F[u] and for the projected free energy F_j(s) in Eq. (8) and Fig. 1. This is confusing and should be disambiguated, e.g., \Phi_j(s) for the free energy.
  2. [Abstract and Introduction] The abstract states that rates are obtained 'from umbrella sampling alone,' but the method also requires a diffusion constant estimate D0 (main text, chignolin section). Please mention this input in the abstract to avoid overstatement.
  3. [Fig. 2 caption] The villin RMSE of 0.21 is said to be limited by the reference (only ≈15 committed folding events). The caption should state the statistical uncertainty of the reference explicitly, since the reader cannot otherwise interpret the quality of the sliced committor.
  4. [Supplement, Eq. (S14)] The held-out quotient used to select hyperparameters is a reasonable model-selection rule, but the text should clarify that the folds are contiguous blocks within states; this is mentioned only in passing. Also, the grid over ε is extensive in M; please state the exact range used for each system in the tables.
  5. [Acknowledgments] There is a typo: 'at the the Gauss Centre' should read 'at the Gauss Centre.'

Circularity Check

0 steps flagged

No significant circularity: the derivation is self-contained; the one acknowledged approximation is not a hidden fit.

full rationale

The central identity Eq. (5) is derived from Green's identity and Cauchy-Schwarz, not from fitting. The only input replacement is Eq. (10), where the flux-weighted fidelity F_j is replaced by the basin-moment difference b_j-a_j; the paper explicitly flags this as the point where the theorems become estimates and quantifies the bias, so it is an acknowledged approximation rather than a fitted parameter renamed as prediction. Hyperparameters are selected by minimizing the held-out cap Eq. (S14), and rates are compared to independent references only after computation. The chignolin non-plateau and high-dimensional substitution are validation concerns, not circularity: no step reduces the central claim to its own inputs by construction. Self-citations are not load-bearing. Therefore score 0.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 1 invented entities

The method introduces one new theoretical object (fidelity) and several hyperparameters selected on data. The exact variational principle is clean, but the practical estimator depends on the domain assumption that equilibrium basin moments approximate flux-weighted fidelities, and on the expressiveness of ridge functions. No new physical entities are postulated.

free parameters (5)
  • Tikhonov ridge ε = selected per system by minimizing Eq. (S14) on held-out folds
    Regularizes the ill-conditioned Gram matrix G; chosen by minimizing the variational quotient on held-out samples, a form of data-driven hyperparameter selection.
  • Direction sampler concentration (µ, α) = AIB9 (0.4,0.2), villin (0.8,0.2), chignolin (0.6,0.4)
    Power-spherical mixture parameters controlling how concentrated directions are around the LDA axis; selected by the same held-out cap criterion.
  • LDA shrinkage ε_LDA = 10^-2 (peptides), 10^-1 (chignolin)
    Ridge added to pooled within-state covariance in Eq. (S11); set by hand.
  • Absorber strength κ = 10^12 (molecular), 10^24 (2D)
    Penalty strength in slice profile Eq. (8). The paper argues the κ→∞ limit is insensitive, but κ is finite and fixed a priori.
  • Diffusion constant D0 (chignolin) = D_Q = 1.3×10^-6 ps^-1, mapped to D0 via Jacobian
    Estimated from umbrella sampling via Var(Q)/τ_int(Q) and the change of variables D0 = D_Q/⟨|∇Q|^2⟩. The rate is proportional to D0, so this sets the overall scale of the rate.
axioms (6)
  • domain assumption Overdamped Langevin dynamics with constant diffusion tensor D (main derivation) or position-dependent D (rate application)
    The variational principle and the flux–fidelity identity are derived for reversible overdamped Langevin dynamics; all applications assume this model.
  • standard math The committor minimizes the Dirichlet energy and satisfies ∇·(ρD∇q)=0 with boundary values
    Standard transition-path theory result, cited to [61-63].
  • standard math Current conservation: ∇·(ρD∇q)=0 in the transition region
    Used to derive the flux–fidelity identity, Eq. (4); true for reversible diffusions.
  • domain assumption Empirical fidelity \hat F_j = b_j − a_j approximates the flux-weighted fidelity F_j
    The central approximation of the method. Validated on 2D benchmarks and AIB9 reactive trajectories, but not exact in high dimensions.
  • domain assumption Ridge-function trial space (one-dimensional projections) is expressive enough to represent the committor
    Assumes committor depends on few collective coordinates (citations [6,66-68]); supported by benchmark results but not guaranteed.
  • domain assumption The isocommittor flux profile plateaus in the transition band, allowing rate read-off via Eq. (S16)
    Requires the stratified flux to be constant over the saddle band; the paper itself reports that this premise fails for chignolin.
invented entities (1)
  • Fidelity functional F[u] no independent evidence
    purpose: Replaces the boundary conditions in the variational principle; equals the flux-weighted boundary average of the trial function difference.
    A new mathematical object defined in Eq. (4). It is not directly measurable and is approximated by equilibrium basin moments; it is a functional, not a physical entity.

pith-pipeline@v1.3.0-daily-deepseek · 31790 in / 10804 out tokens · 102500 ms · 2026-08-04T05:08:54.608410+00:00 · methodology

0 comments
read the original abstract

The committor is the optimal reaction coordinate for a rare transition: it pinpoints the transition state and fixes the rate, and it governs events from protein folding to crystal nucleation. It minimises a Dirichlet energy, whose value at the minimum is the reactive flux, over functions that vanish on the reactant state and equal one on the product state. In high dimensions such a trial space is very hard to build. Here we rewrite the variational principle so that the boundary conditions are replaced by a normalisation of one boundary observable, the fidelity. Any trial function is then admissible, including functions that cannot satisfy the boundary values at all. On this basis we estimate the high-dimensional committor and rates from one-dimensional profiles along projected coordinates, taking as input only pre-existing equilibrium or reweighted configurations, a diffusion constant estimate, and the two state definitions. The optimum has a closed form that is cheap to evaluate. We obtain committors for AIB9 and villin HP-35 in full torsion space, and folding and unfolding rates for chignolin from umbrella sampling alone.

Figures

Figures reproduced from arXiv: 2608.02536 by Magnus Petersen, Roberto Covino, Simon Lichtinger.

Figure 1
Figure 1. Figure 1: FIG. 1. The sliced committor in three steps. (a) Projection directions [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Sliced committor for molecular systems. In both rows the committor is computed in the full sine and cosine torsion [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Chignolin (CLN025) sliced committor, computed in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

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

96 extracted references · 7 linked inside Pith

  1. [1]

    R. Du, V. S. Pande, A. Y. Grosberg, T. Tanaka, and E. S. Shakhnovich, On the transition coordinate for pro- tein folding, The Journal of Chemical Physics108, 334 (1998). 7

    1998

  2. [2]

    A. M. Berezhkovskii and A. Szabo, One-dimensional re- action coordinates for diffusive activated rate processes in many dimensions, The Journal of Chemical Physics 122, 014503 (2005)

    2005

  3. [3]

    W. E and E. Vanden-Eijnden, Transition-Path Theory and Path-Finding Algorithms for the Study of Rare Events, Annual Review of Physical Chemistry61, 391 (2010)

    2010

  4. [4]

    P. G. Bolhuis, D. Chandler, C. Dellago, and P. L. Geissler, Transition Path Sampling: Throwing Ropes over Rough Mountain Passes, in the Dark, Annual re- view of physical chemistry53, 291 (2002)

    2002

  5. [5]

    Hummer, From transition paths to transition states and rate coefficients, The Journal of Chemical Physics 120, 516 (2004)

    G. Hummer, From transition paths to transition states and rate coefficients, The Journal of Chemical Physics 120, 516 (2004)

    2004

  6. [6]

    R. B. Best and G. Hummer, Reaction coordinates and rates from transition paths, Proceedings of the National Academy of Sciences102, 6732 (2005)

    2005

  7. [7]

    S. V. Krivov, Protein Folding Free Energy Landscape along the Committor - the Optimal Folding Coordinate, Journal of Chemical Theory and Computation14, 3418 (2018)

    2018

  8. [8]

    A. M. Berezhkovskii and A. Szabo, Diffusion along the Splitting/Commitment Probability Reaction Coordinate, The Journal of Physical Chemistry B117, 13115 (2013)

    2013

  9. [9]

    Kimura, On the probability of fixation of mutant genes in a population, Genetics47, 713 (1962)

    M. Kimura, On the probability of fixation of mutant genes in a population, Genetics47, 713 (1962)

    1962

  10. [10]

    Lucente, C

    D. Lucente, C. Herbert, and F. Bouchet, Committor Functions for Climate Phenomena at the Predictability Margin: The Example of El Ni˜ no–Southern Oscillation in the Jin and Timmermann Model, Journal of the At- mospheric Sciences79, 2387 (2022)

    2022

  11. [11]

    Finkel, R

    J. Finkel, R. J. Webber, E. P. Gerber, D. S. Abbot, and J. Weare, Learning Forecasts of Rare Stratospheric Tran- sitions from Short Simulations, Monthly Weather Review 149, 3647 (2021)

    2021

  12. [12]

    Vanden-Eijnden, Transition Path Theory, inComputer Simulations in Condensed Matter Systems: From Materi- als to Chemical Biology Volume 1, edited by M

    E. Vanden-Eijnden, Transition Path Theory, inComputer Simulations in Condensed Matter Systems: From Materi- als to Chemical Biology Volume 1, edited by M. Ferrario, G. Ciccotti, and K. Binder (Springer, Berlin, Heidelberg,

  13. [13]

    Roux, Transition rate theory, spectral analysis, and reactive paths, The Journal of Chemical Physics156, 134111 (2022)

    B. Roux, Transition rate theory, spectral analysis, and reactive paths, The Journal of Chemical Physics156, 134111 (2022)

    2022

  14. [14]

    A. M. Berezhkovskii and A. Szabo, Committors, first- passage times, fluxes, Markov states, milestones, and all that, The Journal of Chemical Physics150, 054106 (2019)

    2019

  15. [15]

    Ma and A

    A. Ma and A. R. Dinner, Automatic Method for Iden- tifying Reaction Coordinates in Complex Systems, The Journal of Physical Chemistry B109, 6769 (2005)

    2005

  16. [16]

    Peters and B

    B. Peters and B. L. Trout, Obtaining reaction coordi- nates by likelihood maximization, The Journal of Chem- ical Physics125, 054108 (2006)

    2006

  17. [17]

    Lechner, J

    W. Lechner, J. Rogal, J. Juraszek, B. Ensing, and P. G. Bolhuis, Nonlinear reaction coordinate analysis in the reweighted path ensemble, The Journal of Chemical Physics133, 174110 (2010)

    2010

  18. [18]

    H. Jung, R. Covino, A. Arjun, C. Leitold, C. Del- lago, P. G. Bolhuis, and G. Hummer, Machine-guided path sampling to discover mechanisms of molecular self-organization, Nature Computational Science3, 334 (2023)

    2023

  19. [19]

    N. E. Strand, S. B. Nicholson, H. Vroylandt, and T. R. Gingrich, From high-dimensional committors to reactive insights, The Journal of Chemical Physics161, 224109 (2024)

    2024

  20. [20]

    P. Kang, E. Trizio, and M. Parrinello, Computing the committor with the committor to study the transition state ensemble, Nature Computational Science4, 451 (2024)

    2024

  21. [21]

    Trizio, P

    E. Trizio, P. Kang, and M. Parrinello, Everything every- where all at once: a probability-based enhanced sampling approach to rare events, Nature Computational Science 5, 582 (2025)

    2025

  22. [22]

    G. M. Rotskoff, A. R. Mitchell, and E. Vanden-Eijnden, Active Importance Sampling for Variational Objectives Dominated by Rare Events: Consequences for Optimiza- tion and Generalization, inProceedings of the 2nd Math- ematical and Scientific Machine Learning Conference, Vol. 145 (PMLR, 2022) pp. 757–780

    2022

  23. [23]

    Lazzeri, H

    G. Lazzeri, H. Jung, P. G. Bolhuis, and R. Covino, Molec- ular Free Energies, Rates, and Mechanisms from Data- Efficient Path Sampling Simulations, Journal of Chemical Theory and Computation19, 9060 (2023)

    2023

  24. [24]

    Lazzeri, P

    G. Lazzeri, P. G. Bolhuis, and R. Covino, Optimal Rejection-Free Path Sampling (2025), arXiv:2503.21037 [physics]

    Pith/arXiv arXiv 2025

  25. [25]

    Hartmann and C

    C. Hartmann and C. Sch¨ utte, Efficient rare event simula- tion by optimal nonequilibrium forcing, Journal of Statis- tical Mechanics: Theory and Experiment2012, P11004 (2012)

    2012

  26. [26]

    J. Yuan, A. Shah, C. Bentz, and M. K. Cameron, Opti- mal control for sampling the transition path process and estimating rates, Communications in Nonlinear Science and Numerical Simulation129, 107701 (2024)

    2024

  27. [27]

    A. N. Singh and D. T. Limmer, Splitting probabilities as optimal controllers of rare reactive events, The Journal of Chemical Physics161, 054113 (2024)

    2024

  28. [28]

    Holdijk, Y

    L. Holdijk, Y. Du, F. Hooft, P. Jaini, B. Ensing, and M. Welling, Stochastic Optimal Control for Collective Variable Free Sampling of Molecular Transition Paths (2023), arXiv:2207.02149 [physics, q-bio]

    Pith/arXiv arXiv 2023

  29. [29]

    A. P. Manuel, J. Lambert, and M. T. Woodside, Recon- structing folding energy landscapes from splitting proba- bility analysis of single-molecule trajectories, Proceedings of the National Academy of Sciences112, 7183 (2015)

    2015

  30. [30]

    Covino, M

    R. Covino, M. T. Woodside, G. Hummer, A. Szabo, and P. Cossio, Molecular free energy profiles from force spectroscopy experiments by inversion of observed com- mittors, The Journal of Chemical Physics151, 154115 (2019)

    2019

  31. [31]

    Metzner, C

    P. Metzner, C. Sch¨ utte, and E. Vanden-Eijnden, Illustra- tion of transition path theory on a collection of simple examples, The Journal of Chemical Physics125, 084110 (2006)

    2006

  32. [32]

    No´ e, C

    F. No´ e, C. Sch¨ utte, E. Vanden-Eijnden, L. Reich, and T. R. Weikl, Constructing the equilibrium ensemble of folding pathways from short off-equilibrium simulations, Proceedings of the National Academy of Sciences106, 19011 (2009)

    2009

  33. [33]

    Metzner, C

    P. Metzner, C. Sch¨ utte, and E. Vanden-Eijnden, Transi- tion Path Theory for Markov Jump Processes, Multiscale Modeling & Simulation7, 1192 (2009)

    2009

  34. [34]

    R. R. Coifman and S. Lafon, Diffusion maps, Applied and Computational Harmonic Analysis21, 5 (2006)

    2006

  35. [35]

    Banisch, Z

    R. Banisch, Z. Trstanova, A. Bittracher, S. Klus, and P. Koltai, Diffusion maps tailored to arbitrary non- 8 degenerate Itˆ o processes, Applied and Computational Harmonic Analysis48, 242 (2020)

    2020

  36. [36]

    Evans, M

    L. Evans, M. K. Cameron, and P. Tiwary, Computing committors in collective variables via Mahalanobis diffu- sion maps, Applied and Computational Harmonic Anal- ysis64, 62 (2023)

    2023

  37. [37]

    Lai and J

    R. Lai and J. Lu, Point Cloud Discretization of Fokker– Planck Operators for Committor Functions, Multiscale Modeling & Simulation16, 710 (2018)

    2018

  38. [38]

    Y. Chen, J. Hoskins, Y. Khoo, and M. Lindsey, Commit- tor functions via tensor networks, Journal of Computa- tional Physics472, 111646 (2023)

    2023

  39. [39]

    E. H. Thiede, D. Giannakis, A. R. Dinner, and J. Weare, Galerkin approximation of dynamical quantities using trajectory data, The Journal of Chemical Physics150, 244111 (2019)

    2019

  40. [40]

    Strahan, A

    J. Strahan, A. Antoszewski, C. Lorpaiboon, B. P. Vani, J. Weare, and A. R. Dinner, Long-Time-Scale Predictions from Short-Trajectory Data: A Benchmark Analysis of the Trp-Cage Miniprotein, Journal of Chemical Theory and Computation17, 2948 (2021)

    2021

  41. [41]

    Strahan, S

    J. Strahan, S. C. Guo, C. Lorpaiboon, A. R. Dinner, and J. Weare, Inexact iterative numerical linear algebra for neural network-based spectral estimation and rare-event prediction, The Journal of Chemical Physics159, 014110 (2023)

    2023

  42. [42]

    Aristoff, M

    D. Aristoff, M. Johnson, G. Simpson, and R. J. Web- ber, The fast committor machine: Interpretable predic- tion with kernels, The Journal of Chemical Physics161, 084113 (2024), arXiv:2405.10410

    Pith/arXiv arXiv 2024

  43. [43]

    Roux, String method with swarms-of-trajectories, mean drifts, lag time, and committor, The Journal of Physical Chemistry A125, 7558 (2021)

    B. Roux, String method with swarms-of-trajectories, mean drifts, lag time, and committor, The Journal of Physical Chemistry A125, 7558 (2021)

    2021

  44. [44]

    Z. He, C. Chipot, and B. Roux, Committor-Consistent Variational String Method, The Journal of Physical Chemistry Letters13, 9263 (2022)

    2022

  45. [45]

    Y. Khoo, J. Lu, and L. Ying, Solving for high-dimensional committor functions using artificial neural networks, Re- search in the Mathematical Sciences6, 1 (2019)

    2019

  46. [46]

    Q. Li, B. Lin, and W. Ren, Computing committor func- tions for the study of rare events using deep learning, The Journal of Chemical Physics151, 054112 (2019), arXiv:1906.06285 [physics]

    Pith/arXiv arXiv 2019

  47. [47]

    H. Chen, B. Roux, and C. Chipot, Discovering Reaction Pathways, Slow Variables, and Committor Probabilities with Machine Learning, Journal of Chemical Theory and Computation19, 4414 (2023)

    2023

  48. [48]

    Meg ´ ıas, S

    A. Meg ´ ıas, S. Contreras Arredondo, C. G. Chen, C. Tang, B. Roux, and C. Chipot, Iterative variational learning of committor-consistent transition pathways using artificial neural networks, Nature Computational Science5, 592 (2025)

    2025

  49. [49]

    M. R. Hasyim, C. H. Batton, and K. K. Mandadapu, Supervised learning and the finite-temperature string method for computing committor functions and reac- tion rates, The Journal of Chemical Physics157, 184111 (2022)

    2022

  50. [50]

    Contreras Arredondo, C

    S. Contreras Arredondo, C. Tang, R. A. Talmazan, A. Meg ´ ıas, C. G. Chen, and C. Chipot, Learning the committor without collective variables, Nature Compu- tational Science6, 350 (2026)

    2026

  51. [51]

    Strahan, J

    J. Strahan, J. Finkel, A. R. Dinner, and J. Weare, Pre- dicting rare events using neural networks and short- trajectory data, Journal of Computational Physics488, 112152 (2023)

    2023

  52. [52]

    A. R. Mitchell and G. M. Rotskoff, Committor Guided Estimates of Molecular Transition Rates, Journal of Chemical Theory and Computation20, 9378 (2024)

    2024

  53. [53]

    Pinkus,Ridge Functions, Cambridge Tracts in Math- ematics (Cambridge University Press, Cambridge, 2015)

    A. Pinkus,Ridge Functions, Cambridge Tracts in Math- ematics (Cambridge University Press, Cambridge, 2015)

    2015

  54. [54]

    J. H. Friedman and W. Stuetzle, Projection Pursuit Re- gression, Journal of the American Statistical Association 76, 817 (1981)

    1981

  55. [55]

    Rabin, G

    J. Rabin, G. Peyr´ e, J. Delon, and M. Bernot, Wasser- stein Barycenter and Its Application to Texture Mixing, inScale Space and Variational Methods in Computer Vi- sion, edited by A. M. Bruckstein, B. M. ter Haar Romeny, A. M. Bronstein, and M. M. Bronstein (Springer, Berlin, Heidelberg, 2012) pp. 435–446

    2012

  56. [56]

    Bonneel, J

    N. Bonneel, J. Rabin, G. Peyr´ e, and H. Pfister, Sliced and Radon Wasserstein Barycenters of Measures, Journal of Mathematical Imaging and Vision51, 22 (2015)

    2015

  57. [57]

    Kolouri, K

    S. Kolouri, K. Nadjahi, U. Simsekli, R. Badeau, and G. K. Rohde, Generalized sliced Wasserstein distances, inAdvances in Neural Information Processing Systems, Vol. 32 (2019) arXiv:1902.00434

    Pith/arXiv arXiv 2019

  58. [58]

    Y. Song, S. Garg, J. Shi, and S. Ermon, Sliced score matching: A scalable approach to density and score esti- mation, inProceedings of the 35th Conference on Uncer- tainty in Artificial Intelligence (UAI), PMLR, Vol. 115 (2020) pp. 574–584, arXiv:1905.07088

    Pith/arXiv arXiv 2020

  59. [59]

    G. M. Torrie and J. P. Valleau, Nonphysical sampling dis- tributions in Monte Carlo free-energy estimation: Um- brella sampling, Journal of Computational Physics23, 187 (1977)

    1977

  60. [60]

    Laio and M

    A. Laio and M. Parrinello, Escaping free-energy min- ima, Proceedings of the National Academy of Sciences 99, 12562 (2002)

    2002

  61. [61]

    Bovier, M

    A. Bovier, M. Eckhoff, V. Gayrard, and M. Klein, Metastability in reversible diffusion processes I: Sharp asymptotics for capacities and exit times, Journal of the European Mathematical Society6, 399 (2004)

    2004

  62. [62]

    Bovier and F

    A. Bovier and F. den Hollander,Metastability: A Potential-Theoretic Approach, Grundlehren der mathe- matischen Wissenschaften, Vol. 351 (Springer, 2015)

    2015

  63. [63]

    Leli` evre and G

    T. Leli` evre and G. Stoltz, Partial differential equations and stochastic methods in molecular dynamics, Acta Nu- merica25, 681 (2016)

    2016

  64. [64]

    Lorpaiboon, J

    C. Lorpaiboon, J. Weare, and A. R. Dinner, An exact multiple-time-step variational formulation for the com- mittor and the transition rate, The Journal of Physical Chemistry B130, 155 (2026)

    2026

  65. [65]

    See Supplemental Material at [URL will be inserted by publisher] for the reaction-diffusion slice formulation, weight derivation, direction-sampling schemes, and im- plementation details

  66. [66]

    P. G. Bolhuis, C. Dellago, and D. Chandler, Reaction coordinates of biomolecular isomerization, Proceedings of the National Academy of Sciences97, 5877 (2000)

    2000

  67. [67]

    Bittracher, P

    A. Bittracher, P. Koltai, S. Klus, R. Banisch, M. Dell- nitz, and C. Sch¨ utte, Transition manifolds of complex metastable systems: Theory and data-driven computa- tion of effective dynamics, Journal of Nonlinear Science 28, 471 (2018)

    2018

  68. [68]

    Bittracher, S

    A. Bittracher, S. Klus, B. Hamzi, P. Koltai, and C. Sch¨ utte, Dimensionality reduction of complex metastable systems via kernel embeddings of transition manifolds, Journal of Nonlinear Science31, 3 (2021). 9

    2021

  69. [69]

    R. J. Webber, E. H. Thiede, D. Dow, A. R. Dinner, and J. Weare, Error bounds for dynamical spectral estima- tion, SIAM Journal on Mathematics of Data Science3, 225 (2021)

    2021

  70. [70]

    P´ erez-Hern´ andez, F

    G. P´ erez-Hern´ andez, F. Paul, T. Giorgino, G. De Fab- ritiis, and F. No´ e, Identification of slow molecular order parameters for Markov model construction, The Journal of Chemical Physics139, 015102 (2013)

    2013

  71. [71]

    C. R. Schwantes and V. S. Pande, Improvements in Markov State Model Construction Reveal Many Non- Native Interactions in the Folding of NTL9, Journal of Chemical Theory and Computation9, 2000 (2013)

    2000

  72. [72]

    Tiwary and M

    P. Tiwary and M. Parrinello, From metadynamics to dy- namics, Physical Review Letters111, 230602 (2013)

    2013

  73. [73]

    Honda, T

    S. Honda, T. Akiba, Y. S. Kato, Y. Sawada, M. Sekijima, M. Ishimura, A. Ooishi, H. Watanabe, T. Odahara, and K. Harata, Crystal structure of a ten-amino acid protein, Journal of the American Chemical Society130, 15327 (2008)

    2008

  74. [74]

    M. R. Shirts and J. D. Chodera, Statistically optimal analysis of samples from multiple equilibrium states, The Journal of Chemical Physics129, 124105 (2008)

    2008

  75. [75]

    R. B. Best, G. Hummer, and W. A. Eaton, Native con- tacts determine protein folding mechanisms in atomistic simulations, Proceedings of the National Academy of Sci- ences110, 17874 (2013)

    2013

  76. [76]

    G. Hummer, Position-dependent diffusion coefficients and free energies from Bayesian analysis of equilibrium and replica molecular dynamics simulations, New Jour- nal of Physics7, 34 (2005)

    2005

  77. [77]

    A. M. Berezhkovskii and A. Szabo, Time scale separation leads to position-dependent diffusion along a slow coor- dinate, The Journal of Chemical Physics135, 074108 (2011)

    2011

  78. [78]

    R. B. Best and G. Hummer, Coordinate-dependent dif- fusion in protein folding, Proceedings of the National Academy of Sciences107, 1088 (2010)

    2010

  79. [79]

    Lindorff-Larsen, S

    K. Lindorff-Larsen, S. Piana, R. O. Dror, and D. E. Shaw, How Fast-Folding Proteins Fold, Science334, 517 (2011)

    2011

  80. [80]

    De Cao and W

    N. De Cao and W. Aziz, The power spherical distribution (2020), ICML 2020 Workshop INNF+, arXiv:2006.04437 [stat.ML]

    Pith/arXiv arXiv 2020

Showing first 80 references.