Pith. sign in

REVIEW 2 major objections 4 minor 52 references

Coarse-grained Markov state models need two transition matrices — one equilibrium, one source-sink — to compute unbiased observables at any lag time.

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

2026-08-01 13:23 UTC pith:PFJ5HEUI

load-bearing objection A careful, useful perspective that proves the two-matrix principle for unbiased MSM observables, but the practical claim leans on RiteWeight's unproven convergence here. the 2 major comments →

arxiv 2607.19452 v2 pith:PFJ5HEUI submitted 2026-07-21 cond-mat.stat-mech physics.chem-phq-bio.BM

Markov state models revisited: Principles and algorithms for unbiased observables

David Aristoff , Robert J. Webber , Daniel M. Zuckerman This is my paper
classification cond-mat.stat-mech physics.chem-phq-bio.BM
keywords Markov state modelsnonequilibrium steady statecoarse-grained observablesmean first-passage timecommittortrajectory reweightingmolecular dynamicssystematic bias
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.

This paper argues that the standard Markov state model (MSM) recipe — build one transition matrix from available trajectory data and read off every observable from it — is systematically biased, even in the infinite-data limit. The unbiased alternative is to construct two ensemble-matched transition matrices: one from the equilibrium distribution and one from the A-to-B nonequilibrium steady state. When each matrix is built from the correct stationary within-cluster sampling distribution, coarse-grained stationary probabilities, fluxes, mean first-passage times, and committors are all unbiased for any fixed lag time and any fixed coarse-graining. The paper supports the claim with a fully solvable eight-state model and exact error formulas that show how transition-matrix errors are amplified by slow relaxation modes.

Core claim

The central claim is that unbiased coarse-grained observables require matching the transition matrix to the ensemble that defines the observable. The equilibrium matrix T^equil supplies equilibrium stationary probabilities; the A→B source-sink matrix T^{A→B} supplies nonequilibrium stationary probabilities, the mean first-passage time via m^{A→B} = τ/(Σ_{i∈B} π_i^{A→B}) − τ, and the cluster-to-cluster flux via Φ_{ij} = π_i T_{ij}; the committor is the normalized ratio q_i^A = C · π_i^{A→B}/π_i^equil. A key point is that the traditional route — impose source-sink boundaries after constructing the equilibrium matrix — is theoretically flawed even with infinite equilibrium sampling, because the

What carries the argument

The machinery is a pair of ensemble-matched coarse-grained transition matrices T^X_{ij} = ∫ ρ^X(x'|i) ∫ p^X_τ(x|x') 1_j(x) dx dx', where ρ^X is the stationary density conditioned on the cluster. For X = equil this yields equilibrium populations; for X = A→B, with recycling via the reactive entrance distribution ρ^A, it yields the NESS populations, mean first-passage time, and flux. Two identities carry the estimates: the discrete-time Hill relation m^{A→B} = τ/(Σ_{i∈B} π_i^{A→B}) − τ, and the committor likelihood-ratio identity q_i^A = (π^{A→B}_i/π^equil_i) · (π^equil_a/π^{A→B}_a) for a source cluster a. The error analysis centers on fundamental matrices that amplify any discrepancy between

Load-bearing premise

The whole A→B machinery assumes the reactive entrance distribution ρ^A (the distribution of starting points in the source state) is available and the recycling process is ergodic; if that input is misspecified, the NESS matrix and every nonequilibrium observable built from it are biased.

What would settle it

In the paper's own eight-state model, compute the stationary vector of the traditional coarse-grained A→B matrix (constructed by imposing recycling on the equilibrium matrix) and compare it with the coarse-grained microscopic NESS vector. If the two ever coincide for any lag time — or if the traditional MFPT from the first-step relation matches the Hill-relation MFPT — the paper's central bias claim is wrong.

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

If this is right

  • Even with infinite equilibrium trajectory data, traditional MSM estimates of rates, committors, and nonequilibrium populations keep a systematic bias that no amount of data can remove.
  • Users can choose lag times based on the physics they want to resolve instead of the need to enforce Markovianity, since unbiasedness holds at any fixed lag time.
  • Mean first-passage times and committors can be obtained from stationary probabilities of two ensembles rather than from first-step relations that assume coarse-grained Markovianity.
  • Trajectory reweighting that drives within-cluster sampling toward the target stationary distribution becomes a core step, not an optional correction.
  • The error formulas give a diagnostic: transition-matrix errors that feed slowly relaxing or frequently visited states are the ones that most corrupt stationary probabilities and kinetics.

Where Pith is reading between the lines

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

  • This suggests a general design principle: every coarse-grained observable should be paired with the ensemble whose stationary distribution defines it; the same matched-matrix logic could be exported to other path observables such as transition-path times or B→A fluxes.
  • A testable extension is to benchmark the two-matrix estimators against long direct simulations on systems with known metastable states, using short lag times where traditional MSMs are most biased.
  • The framework's practicality hinges on approximating the reactive entrance distribution ρ^A; an editor's reading is that the approach would benefit from algorithms that estimate ρ^A jointly with the NESS matrix rather than assuming it is given.

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

2 major / 4 minor

Summary. This paper argues that the standard single-matrix Markov state model (MSM) workflow produces systematically biased coarse-grained observables, even in the infinite-data limit, because different observables are naturally associated with different ensembles. The authors propose replacing the single matrix by two transition matrices: an equilibrium matrix T^equil and an A→B nonequilibrium steady-state matrix T^{A→B}, each built from the corresponding stationary within-cluster distribution. They illustrate the core bias with an exactly solvable 8-state toy model (Sec. 2), give continuous-state definitions of the A→B NESS and committor (Sec. 3), present unbiased formulas for stationary probabilities, fluxes, MFPT, and committors (Sec. 4), discuss trajectory reweighting via RiteWeight (Sec. 5), and derive exact error expressions for standard MSM estimates (Sec. 6). The paper is explicitly a perspective, with the main theoretical contribution being the identification of which ensemble-matched matrix is required for each observable.

Significance. If the claims are taken as conditional on having the exact NESS density, the paper makes a valuable and largely correct conceptual point: the coarse-grained dynamics need not be Markovian for certain observables to be estimated without bias, provided the transition matrix is constructed from the correct stationary within-cluster distribution. The 8-state toy model is internally consistent and clearly demonstrates the failure of the equilibrium-weighted matrix for NESS observables. The Kac/Hill relation for the MFPT and the likelihood-ratio committor formula are clean and useful; Propositions 6.1–6.3 are correct and give interpretable error decompositions. The paper’s main weakness is that the practical route to the exact A→B NESS density, RiteWeight, is only cited to a separate paper, so the headline claim of “how to obtain” unbiased observables is stronger than what is established here.

major comments (2)
  1. [Sec. 5.3 / Sec. 4.1 / Abstract] The abstract claims that the paper “shows how to obtain unbiased coarse-grained observables” at any fixed lag time and coarse-graining. This is stronger than what is established. The matrix T^{A→B} in Eq. (40) is constructed from the exact NESS density π^{A→B}, and Sec. 4.1 explicitly concedes that “this approach for calculating stationary probabilities may seem circular.” The practical algorithm intended to resolve the circularity, RiteWeight in Eqs. (63)–(64), is asserted to be asymptotically unbiased only by citation to [20]; no theorem, sufficient conditions, or proof are given here. Any misspecification or non-convergence in estimating π^{A→B} propagates directly into the nonequilibrium formulas, including the stationary probabilities, Eq. (47), the committor, Eq. (51), and the flux expressions. The revision should either state the RiteWeight convergence result precisely, with expli
  2. [Sec. 3.3, Eqs. (22)–(23)] The committor estimator (51) rests entirely on the identity q^A(x) = π^{A→B} π^{A→B}(x)/π^{equil}(x), which is asserted in Eq. (22) with citations but not derived in the text. This relation is not an immediate consequence of the stationarity condition (20); it connects the stationary density of the recycling process (19), whose source ρ^A is itself a dynamical first-passage object, to the equilibrium density and the committor. Since the committor is one of the four headline observables in Table 1, the revision should provide a derivation or a precise statement of the theorem being invoked, including the detailed-balance or momentum-flip hypotheses and the treatment of the sink region B. As written, a reader cannot verify the central identity from the manuscript alone.
minor comments (4)
  1. [Throughout] Typos: “whatnotto do” in Sec. 1, “approriate” in Sec. 4.1, “tajectories” in Sec. 5.3. Please copyedit.
  2. [Sec. 5.3, Eqs. (63)–(64)] The algorithm description says the cluster definition is randomly changed at each iteration, but the displayed update equations do not reflect the random re-clustering. A reader must consult [20] to understand the procedure; since RiteWeight is load-bearing for the practical claims, a brief specification or a precise pointer to the theorem would help.
  3. [Sec. 6.2, Eq. (73)] In Proposition 6.2, the statement defines \tilde m but not \tilde m^B; the proof uses both. Please define both vectors in the proposition, noting that \tilde m is the solution of the first-step relation for the computed matrix T and \tilde m^B is the exact MFPT vector for T^{A→B}.
  4. [Sec. 4.2, Eq. (44)] For the A→B ensemble, the flux Φ^{A→B}_{ij} includes the recycling step from B to A because p^{A→B}_τ in Eq. (19) resets points in B to ρ^A. This should be stated explicitly so that the mechanistic flux is not confused with a reactive flux that excludes the source-sink recycling.

Circularity Check

1 steps flagged

Acknowledged fixed-point identity: NESS stationary probabilities are presupposed by the construction of T^{A→B}, not predicted; practical unbiasedness is delegated to the self-cited RiteWeight algorithm.

specific steps
  1. self definitional [Sec. 4.1, Eqs. (40)–(43)]
    "We form the A→B NESS MSM with elements given by T^{A→B}_{ij} = ∫ π^{A→B}(x'|i) ∫ p^{A→B}_τ(x|x') I_j(x) dx dx'. ... Similar to (38), the stationary vector of the transition matrix T^{A→B}_{ij} yields the probabilities π^{A→B}_i, ... This approach for calculating stationary probabilities may seem circular, but it underpins important practical algorithms as we will see in Sec. 5."

    Equation (40) constructs the transition matrix from the NESS conditional density π^{A→B}(x|i); Eq. (43) then recovers the cluster probabilities π^{A→B}_i as its stationary vector. The recovery is an identity: a matrix built from a target distribution necessarily has that distribution's cluster sums as its stationary vector. The paper does not derive π^{A→B} from data in this section; it is an input. Consequently, the unbiasedness claims for MFPT (Eq. 47) and committor (Eq. 51) are conditional on having the exact π^{A→B} already. The paper's own text concedes the apparent circularity.

full rationale

The dominant circularity is the explicitly flagged fixed-point structure in Sec. 4.1: T^{A→B} is defined as the conditional NESS average of the microscopic dynamics, so its stationary vector is the coarse-grained NESS distribution by construction. This is not a prediction in the usual sense, and the paper acknowledges it, but it means the 'unbiased' NESS observables (stationary probabilities, MFPT, committor) are functionals of an input density rather than outputs of a data-driven derivation. The paper's independence from this fixed point lies in its counterexample analysis (Sec. 2) showing traditional single-matrix MSMs are biased, and in the error analysis (Sec. 6), which does not rely on the fixed-point identity. The practical estimation of π^{A→B} is delegated to RiteWeight, whose asymptotic unbiasedness is cited from [20]—part of the same research program (cf. refs. [1], [26], [30], [31])—without proof in this manuscript. However, the central theoretical conditional claim is internally consistent, and the self-citations are for background and algorithm delegation rather than for a uniqueness theorem. This warrants a moderate score: partial circularity by construction, but substantial independent content in the bias analysis and error formulas.

Axiom & Free-Parameter Ledger

1 free parameters · 5 axioms · 0 invented entities

The framework introduces no new physical entities; the A→B NESS density and its associated transition matrix are mathematical objects defined from the microscopic dynamics. The primary concession is that the NESS distribution is an input to the unbiased estimators, along with the usual background assumptions (Markovianity in full phase space, detailed balance). The lag time τ is a free modeling choice held fixed throughout.

free parameters (1)
  • lag time τ = user-chosen
    All observables depend on the discretization lag time; the paper's claims hold for any fixed τ, but the value of the MFPT and flux depend on it (Secs. 3.4, 4.3). It is a modeling choice, not fitted to data.
axioms (5)
  • domain assumption Microscopic dynamics in full phase space are Markovian
    Sec. 3, before Eq. (15): 'Because the system is Markovian in the full phase space, only a single prior phase point x' is needed to specify the distribution of outcomes.' This justifies treating transition densities as the fundamental object.
  • domain assumption Equilibrium density satisfies detailed balance (or momentum-flip detailed balance)
    Eq. (17) and surrounding text; used in Sec. 3.3 to derive the committor likelihood ratio (Eqs. 22–23) and in Sec. 3.2 to relate equilibrium and NESS ensembles.
  • domain assumption The A→B recycling process is stationary and ergodic, with a well-defined reactive entrance distribution ρ^A (EqSurf)
    Sec. 3.2 constructs p^{A→B} from ρ^A in Eq. (19); Sec. 3.4's Kac return-time argument (Eqs. 26–29) requires stationarity and ergodicity of the recycled chain.
  • ad hoc to paper The A→B NESS density π^{A→B}(x) is available or can be estimated without bias
    Eqs. (11) and (40) define the A→B transition matrix using π^{A→B}(x|i). The unbiasedness of all nonequilibrium observables depends on this input; the paper defers estimation to RiteWeight [20] without proving convergence here.
  • domain assumption Coarse clusters are non-overlapping and cover phase space
    Sec. 4 and Eq. (34): indicator functions I_j are used; Eq. (39) relies on the partition of unity to relate conditional and global densities.

pith-pipeline@v1.3.0-alltime-deepseek · 20242 in / 15961 out tokens · 137184 ms · 2026-08-01T13:23:18.358684+00:00 · methodology

0 comments
read the original abstract

Markov state models (MSMs) have become ubiquitous tools for analyzing molecular dynamics (MD) simulations because of their simple, powerful premise: although complete MD sampling may be impossible, the MSM can "stitch together" transition probabilities derived from local sampling to provide a global picture of kinetics and mechanisms. In the standard MSM framework, the available MD data is organized into a single transition matrix, which is then used to estimate all observables at a lag time chosen so the coarse-grained dynamics are approximately Markovian. This approach leads to avoidable model bias and motivates long lag times that obscure short-timescale processes of interest. In contrast, this paper shows how to obtain unbiased coarse-grained observables at any fixed lag time and for any fixed coarse-graining in the limit of infinite, properly weighted data. The central idea is to replace the single-matrix framework with two transition matrices -- one representing equilibrium dynamics and another representing source-sink recycling dynamics -- and use the correct matrix or matrices to estimate the matched dynamical observables.

Figures

Figures reproduced from arXiv: 2607.19452 by Daniel M. Zuckerman, David Aristoff, Robert J. Webber.

Figure 1
Figure 1. Figure 1: Markov state modeling pipeline and the corresponding sources of error. This perspective focuses [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic rough energy landscape and clusters of an MSM. A biomacromolecule consists of [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Decomposition of an equilibrium trajectory ensemble into two nonequilibrium steady-state en [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Long trajectories in the equilibrium and [PITH_FULL_IMAGE:figures/full_fig_p010_4.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

52 extracted references · 29 canonical work pages · 2 internal anchors

  1. [1]

    Zuckerman.bRiteWeight: Randomized iterative reweighting for biased trajectory data

    David Aristoff and Daniel M. Zuckerman.bRiteWeight: Randomized iterative reweighting for biased trajectory data. ChemRxiv. 2026.doi:10.26434/chemrxiv.15001675/v1

    work page doi:10.26434/chemrxiv.15001675/v1 2026

  2. [2]

    On the removal of initial state bias from simula- tion data

    Marco Bacci, Amedeo Caflisch, and Andreas Vitalis. “On the removal of initial state bias from simula- tion data”. In:The Journal of Chemical Physics150.10 (2019), p. 104105.doi:10.1063/1.5063556

    work page doi:10.1063/1.5063556 2019

  3. [3]

    On the Hill relation and the mean reaction time for metastable processes

    Manon Baudel, Arnaud Guyader, and Tony Leli` evre. “On the Hill relation and the mean reaction time for metastable processes”. In:Stochastic Processes and their Applications155 (2023), pp. 393–436. doi:10.1016/j.spa.2022.10.014

    work page doi:10.1016/j.spa.2022.10.014 2023

  4. [4]

    Beyond microscopic reversibility: Are observable nonequi- librium processes precisely reversible?

    Divesh Bhatt and Daniel M. Zuckerman. “Beyond microscopic reversibility: Are observable nonequi- librium processes precisely reversible?” In:Journal of Chemical Theory and Computation7.8 (2011), pp. 2520–2527.doi:10.1021/ct200086k. 22

    work page doi:10.1021/ct200086k 2011

  5. [5]

    On the relation between projections of the reweighted path ensemble

    Peter G. Bolhuis and Wolfgang Lechner. “On the relation between projections of the reweighted path ensemble”. In:Journal of Statistical Physics145.4 (2011), pp. 841–859.doi:10.1007/s10955-011- 0324-6

    work page doi:10.1007/s10955-011- 2011

  6. [6]

    New York: Oxford University Press, 1987.isbn: 978-0-19-504277-1.url:https://openlibrary.org/books/OL2724292M/Introduction_ to_modern_statistical_mechanics

    David Chandler.Introduction to Modern Statistical Mechanics. New York: Oxford University Press, 1987.isbn: 978-0-19-504277-1.url:https://openlibrary.org/books/OL2724292M/Introduction_ to_modern_statistical_mechanics

    1987

  7. [7]

    Markov state models of biomolecular conformational dynamics

    John D. Chodera and Frank No´ e. “Markov state models of biomolecular conformational dynamics”. In:Current Opinion in Structural Biology25 (2014), pp. 135–144.doi:10.1016/j.sbi.2014.04.002

    work page doi:10.1016/j.sbi.2014.04.002 2014

  8. [8]

    Probability distributions of molecular observables computed from Markov models. II. Uncertainties in observables and their time evolution

    John D. Chodera and Frank No´ e. “Probability distributions of molecular observables computed from Markov models. II. Uncertainties in observables and their time evolution”. In:The Journal of Chemical Physics133.10 (2010), p. 105102.doi:10.1063/1.3463406

    work page doi:10.1063/1.3463406 2010

  9. [9]

    Accelerated estimation of long-timescale kinetics from weighted ensemble simulation via non-Markovian “microbin

    Jeremy Copperman and Daniel M. Zuckerman. “Accelerated estimation of long-timescale kinetics from weighted ensemble simulation via non-Markovian “microbin” analysis”. In:Journal of Chemical Theory and Computation16.11 (2020), pp. 6763–6775.doi:10.1021/acs.jctc.0c00273

    work page doi:10.1021/acs.jctc.0c00273 2020

  10. [10]

    Analysis of the accelerated weighted ensemble methodology

    Ronan Costaouec et al. “Analysis of the accelerated weighted ensemble methodology”. In:Discrete and Continuous Dynamical Systems(2013), pp. 171–181.doi:10.3934/proc.2013.2013.171

    arXiv 2013

  11. [11]

    Reaction path study of conformational transitions in flexible systems: Applications to peptides

    Ryszard Czerminski and Ron Elber. “Reaction path study of conformational transitions in flexible systems: Applications to peptides”. In:The Journal of Chemical Physics92.9 (1990), pp. 5580–5601. doi:10.1063/1.458491

    work page doi:10.1063/1.458491 1990

  12. [12]

    Separating forward and backward pathways in nonequilibrium umbrella sampling

    Alex Dickson, Aryeh Warmflash, and Aaron R. Dinner. “Separating forward and backward pathways in nonequilibrium umbrella sampling”. In:The Journal of Chemical Physics131.15 (2009), p. 154104. doi:10.1063/1.3244561

    work page doi:10.1063/1.3244561 2009

  13. [13]

    Gardiner.Handbook of Stochastic Methods for Physics, Chemistry and the Natural Sciences

    Crispin W. Gardiner.Handbook of Stochastic Methods for Physics, Chemistry and the Natural Sciences. 2nd ed. Berlin: Springer-Verlag, 1985.doi:10.1007/978-3-662-02452-2

    doi:10.1007/978-3-662-02452-2 1985

  14. [14]

    Optimized Markov state models for metastable systems

    Enrico Guarnera and Eric Vanden-Eijnden. “Optimized Markov state models for metastable systems”. In:The Journal of Chemical Physics145.2 (2016), p. 024102.doi:10.1063/1.4954769

    work page doi:10.1063/1.4954769 2016

  15. [15]

    MSMBuilder: Statistical models for biomolecular dynamics

    Matthew P. Harrigan et al. “MSMBuilder: Statistical models for biomolecular dynamics”. In:Biophys- ical Journal112.1 (2017), pp. 10–15.doi:10.1016/j.bpj.2016.10.042

    work page doi:10.1016/j.bpj.2016.10.042 2017

  16. [16]

    Hill.Free Energy Transduction and Biochemical Cycle Kinetics

    Terrell L. Hill.Free Energy Transduction and Biochemical Cycle Kinetics. New York: Springer-Verlag, 1989.doi:10.1007/978-1-4612-3558-3

    work page doi:10.1007/978-1-4612-3558-3 1989

  17. [17]

    Deeptime: A Python library for machine learning dynamical models from time series data

    Moritz Hoffmann et al. “Deeptime: A Python library for machine learning dynamical models from time series data”. In:Machine Learning: Science and Technology3.1 (2021), p. 015009.doi:10.1088/2632- 2153/ac3de0

    doi:10.1088/2632- 2021

  18. [18]

    Markov state models: From an art to a science

    Brooke E. Husic and Vijay S. Pande. “Markov state models: From an art to a science”. In:Journal of the American Chemical Society140.7 (2018), pp. 2386–2396.doi:10.1021/jacs.7b12191

    doi:10.1021/jacs.7b12191 2018

  19. [19]

    Optimized parameter selection reveals trends in Markov state models for protein folding

    Brooke E. Husic et al. “Optimized parameter selection reveals trends in Markov state models for protein folding”. In:The Journal of Chemical Physics145.19 (2016), p. 194103.doi:10.1063/1.4967809

    work page doi:10.1063/1.4967809 2016

  20. [20]

    Randomized iterative trajectory reweighting for steady-state distributions without discretization error

    Sagar Kania et al. “Randomized iterative trajectory reweighting for steady-state distributions without discretization error”. In:Proceedings of the National Academy of Sciences of the United States of America123.19 (2026), e2529246123.doi:10.1073/pnas.2529246123

    work page doi:10.1073/pnas.2529246123 2026

  21. [21]

    Uncertainties in Markov state models of small proteins

    Nicolai Kozlowski and Helmut Grubm¨ uller. “Uncertainties in Markov state models of small proteins”. In:Journal of Chemical Theory and Computation19.16 (2023), pp. 5516–5524.doi:10.1021/acs. jctc.3c00372

    doi:10.1021/acs 2023

  22. [22]

    Benedict Leimkuhler and Charles Matthews.Molecular Dynamics: With Deterministic and Stochastic Numerical Methods. Vol. 39. Interdisciplinary Applied Mathematics. Springer, 2015.doi:10.1007/ 978-3-319-16375-8

    2015

  23. [23]

    Probability distributions of molecular observables computed from Markov models

    Frank No´ e. “Probability distributions of molecular observables computed from Markov models”. In: The Journal of Chemical Physics128.24 (2008), p. 244103.doi:10.1063/1.2916718. 23

    work page doi:10.1063/1.2916718 2008

  24. [24]

    J. R. Norris.Markov Chains. Vol. 2. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 1997.doi:10.1017/CBO9780511810633

    doi:10.1017/cbo9780511810633 1997

  25. [25]

    Markov state models from short non-equilibrium simulations—Analysis and cor- rection of estimation bias

    Feliks N¨ uske et al. “Markov state models from short non-equilibrium simulations—Analysis and cor- rection of estimation bias”. In:The Journal of Chemical Physics146.9 (2017), p. 094104.doi:10. 1063/1.4976518

    2017

  26. [26]

    Zuckerman.Regularized RiteWeight for sparse trajectory data: A smoothed stationary reweighting framework

    Lisa Otten, David Aristoff, and Daniel M. Zuckerman.Regularized RiteWeight for sparse trajectory data: A smoothed stationary reweighting framework. ChemRxiv. 2026.doi:10 . 26434 / chemrxiv . 15001337/v1

    2026

  27. [27]

    Everything you wanted to know about Markov state models but were afraid to ask

    Vijay S. Pande, Kyle Beauchamp, and Gregory R. Bowman. “Everything you wanted to know about Markov state models but were afraid to ask”. In:Methods52.1 (2010), pp. 99–105.doi:10.1016/j. ymeth.2010.06.002

    doi:10.1016/j 2010

  28. [28]

    Markov models of molecular kinetics: Generation and validation

    Jan-Hendrik Prinz et al. “Markov models of molecular kinetics: Generation and validation”. In:The Journal of Chemical Physics134.17 (2011), p. 174105.doi:10.1063/1.3565032

    doi:10.1063/1.3565032 2011

  29. [29]

    Nicolas Privault.Understanding Markov Chains: Examples and Applications. 2nd ed. Springer Under- graduate Mathematics Series. Springer, 2018.doi:10.1007/978-981-13-0659-4

    doi:10.1007/978-981-13-0659-4 2018

  30. [30]

    Iterative trajectory reweighting for estimation of equilibrium and non-equilibrium observables

    John D. Russo, Jeremy Copperman, and Daniel M. Zuckerman.Iterative trajectory reweighting for estimation of equilibrium and non-equilibrium observables. arXiv. 2020.doi:10.48550/arXiv.2006. 09451. arXiv:2006.09451

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2006.09451 2020

  31. [31]

    Unbiased estimation of equilibrium, rates, and committors from Markov state model analysis

    John D. Russo et al.Unbiased estimation of equilibrium, rates, and committors from Markov state model analysis. arXiv. 2021.doi:10.48550/arXiv.2105.13402. arXiv:2105.13402

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2105.13402 2021

  32. [32]

    On the approximation quality of Markov state mod- els

    Marco Sarich, Frank No´ e, and Christof Sch¨ utte. “On the approximation quality of Markov state mod- els”. In:Multiscale Modeling & Simulation8.4 (2010), pp. 1154–1177.doi:10.1137/090764049

    work page doi:10.1137/090764049 2010

  33. [33]

    Equilibrium distribution from distributed computing (simula- tions of protein folding)

    Riccardo Scalco and Amedeo Caflisch. “Equilibrium distribution from distributed computing (simula- tions of protein folding)”. In:The Journal of Physical Chemistry B115.19 (2011), pp. 6358–6365.doi: 10.1021/jp2014918

    work page doi:10.1021/jp2014918 2011

  34. [34]

    PyEMMA 2: A software package for estimation, validation, and analysis of Markov models

    Martin K. Scherer et al. “PyEMMA 2: A software package for estimation, validation, and analysis of Markov models”. In:Journal of Chemical Theory and Computation11.11 (2015), pp. 5525–5542.doi: 10.1021/acs.jctc.5b00743

    doi:10.1021/acs.jctc.5b00743 2015

  35. [35]

    Identifying mechanistically distinct pathways in kinetic tran- sition networks

    Daniel J. Sharpe and David J. Wales. “Identifying mechanistically distinct pathways in kinetic tran- sition networks”. In:The Journal of Chemical Physics151.12 (2019), p. 124101.doi:10.1063/1. 5111939

    doi:10.1063/1 2019

  36. [36]

    Error analysis and efficient sampling in Markovian state models for molecular dynamics

    Nina Singhal and Vijay S. Pande. “Error analysis and efficient sampling in Markovian state models for molecular dynamics”. In:The Journal of Chemical Physics123.20 (2005), p. 204909.doi:10.1063/ 1.2116947

    2005

  37. [37]

    Accurate estimation of protein folding and unfolding times: Beyond Markov state models

    Ernesto Su´ arez, Joshua L. Adelman, and Daniel M. Zuckerman. “Accurate estimation of protein folding and unfolding times: Beyond Markov state models”. In:Journal of Chemical Theory and Computation 12.8 (2016), pp. 3473–3481.doi:10.1021/acs.jctc.6b00339

    work page doi:10.1021/acs.jctc.6b00339 2016

  38. [38]

    Simultaneous computation of dynamical and equilibrium information using a weighted ensemble of trajectories

    Ernesto Su´ arez et al. “Simultaneous computation of dynamical and equilibrium information using a weighted ensemble of trajectories”. In:Journal of Chemical Theory and Computation10.7 (2014), pp. 2658–2667.doi:10.1021/ct401065r

    doi:10.1021/ct401065r 2014

  39. [39]

    What Markov state models can and cannot do: Correlation versus path-based observables in protein-folding models

    Ernesto Su´ arez et al. “What Markov state models can and cannot do: Correlation versus path-based observables in protein-folding models”. In:Journal of Chemical Theory and Computation17.5 (2021), pp. 3119–3133.doi:10.1021/acs.jctc.0c01154

    work page doi:10.1021/acs.jctc.0c01154 2021

  40. [40]

    Describing protein folding kinetics by molecular dynamics simulations. 1. Theory

    William C. Swope, Jed W. Pitera, and Frank Suits. “Describing protein folding kinetics by molecular dynamics simulations. 1. Theory”. In:The Journal of Physical Chemistry B108.21 (2004), pp. 6571– 6581.doi:10.1021/jp037421y

    work page doi:10.1021/jp037421y 2004

  41. [41]

    Describing protein folding kinetics by molecular dynamics simulations. 2. Example applications to alanine dipeptide and aβ-hairpin peptide

    William C. Swope et al. “Describing protein folding kinetics by molecular dynamics simulations. 2. Example applications to alanine dipeptide and aβ-hairpin peptide”. In:The Journal of Physical Chemistry B108.21 (2004), pp. 6582–6594.doi:10.1021/jp037422q. 24

    work page doi:10.1021/jp037422q 2004

  42. [42]

    Estimation and uncertainty of reversible Markov models

    Benjamin Trendelkamp-Schroer et al. “Estimation and uncertainty of reversible Markov models”. In: The Journal of Chemical Physics143.17 (2015), p. 174101.doi:10.1063/1.4934536

    doi:10.1063/1.4934536 2015

  43. [43]

    Error breakdown and sensitivity analysis of dynamical quantities in Markov state models

    Yehor Tuchkov et al. “Error breakdown and sensitivity analysis of dynamical quantities in Markov state models”. In:Journal of Chemical Theory and Computation21.23 (2025), pp. 12304–12316.doi: 10.1021/acs.jctc.5c01143

    work page doi:10.1021/acs.jctc.5c01143 2025

  44. [44]

    N. G. van Kampen.Stochastic Processes in Physics and Chemistry. 3rd ed. Elsevier, 2007.doi:10. 1016/B978-0-444-52965-7.X5000-4

    2007

  45. [45]

    Exact rate calculations by trajectory parallelization and tilting

    Eric Vanden-Eijnden and Maddalena Venturoli. “Exact rate calculations by trajectory parallelization and tilting”. In:The Journal of Chemical Physics131.4 (2009), p. 044120.doi:10.1063/1.3180821

    doi:10.1063/1.3180821 2009

  46. [46]

    Exploring energy landscapes

    David J. Wales. “Exploring energy landscapes”. In:Annual Review of Physical Chemistry69 (2018), pp. 401–425.doi:10.1146/annurev-physchem-050317-021219

    work page doi:10.1146/annurev-physchem-050317-021219 2018

  47. [47]

    Adaptive Markov state model estimation using short reseeding trajectories

    Hongbin Wan and Vincent A. Voelz. “Adaptive Markov state model estimation using short reseeding trajectories”. In:The Journal of Chemical Physics152.2 (2020), p. 024103.doi:10.1063/1.5142457

    work page doi:10.1063/1.5142457 2020

  48. [48]

    Variational Koopman models: Slow collective variables and molecular kinetics from short off-equilibrium simulations

    Hao Wu et al. “Variational Koopman models: Slow collective variables and molecular kinetics from short off-equilibrium simulations”. In:The Journal of Chemical Physics146.15 (2017), p. 154104.doi: 10.1063/1.4979344

    doi:10.1063/1.4979344 2017

  49. [49]

    Zuckerman.“Proof ” of the Hill Relation Between Probability Flux and Mean First-Passage Time

    Daniel M. Zuckerman.“Proof ” of the Hill Relation Between Probability Flux and Mean First-Passage Time. Statistical Biophysics Blog. 2015.doi:10.6083/M4Q81CKN

    work page doi:10.6083/m4q81ckn 2015

  50. [50]

    Zuckerman.Counting is not enough: A weakness of MSMs inherited by RiteWeight

    Daniel M. Zuckerman.Counting is not enough: A weakness of MSMs inherited by RiteWeight. Statisti- cal Biophysics Blog. June 22, 2026.url:https://statisticalbiophysicsblog.org/?p=566(visited on 07/22/2026)

    2026

  51. [51]

    Zuckerman.Statistical Physics of Biomolecules: An Introduction

    Daniel M. Zuckerman.Statistical Physics of Biomolecules: An Introduction. CRC Press, 2010.doi: 10.1201/b18849

    work page doi:10.1201/b18849 2010

  52. [52]

    A gentle introduction to the non-equilibrium physics of trajectories: Theory, algorithms, and biomolecular applications

    Daniel M. Zuckerman and John D. Russo. “A gentle introduction to the non-equilibrium physics of trajectories: Theory, algorithms, and biomolecular applications”. In:American Journal of Physics89.11 (2021), pp. 1048–1061.doi:10.1119/10.0005603. 25

    doi:10.1119/10.0005603 2021