{"id":"9a43d848-9d21-4a6e-be10-c977fc1ec5a0","arxiv_id":"2501.04512","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For non-Markovian dynamics, the maximal transition-path probability is non-monotonic in memory time and can exceed 1/2, invalidating 1/2 as a universal reaction-coordinate quality benchmark.","lead":"For dynamics with memory friction, the maximum transition-path probability can rise above the Markovian benchmark of 1/2 and depends non-monotonically on memory time, so 1/2 is not a reliable reaction-coordinate quality test. The authors combine Grote-Hynes rate theory with generalized Langevin simulations to show this in a double-well model.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The geometric-series derivation of Eq. (4) assumes independent recrossing attempts with constant success probability κ; velocity/memory correlations could invalidate the analytical link between max[p(TP|x)] and the Grote-Hynes transmission coefficient, leaving the simulated >1/2 effect as the main…","rationale":"The central claim has two pillars: direct GLE simulations showing max[p(TP|x)] > 1/2 and eventually approaching 1 for long memory, and an analytical estimate Eq. (4) linking this maximum to the Grote-Hynes transmission coefficient. The reader correctly identifies the geometric-series model in SM V as the weakest assumption of the analytical pillar. This is the most load-bearing concern because the analytical estimate provides the quantitative interpretation of the simulation data, including the extrapolation to the long-memory limit. If the independence assumption fails, Eq. (4) is only heuristic, and the paper must rely on the simulations alone. The simulations appear clear and the qualitative conclusion likely survives, which is why the reader's CONDITIONAL verdict remains appropriate rather than a harsher judgment. The related finite-mass issue is real but explicitly acknowledged by the authors in SM IV; it further supports keeping the verdict conditional, but does not by itself overturn the central claim. The proposed direct measurement of velocity-direction commitors would settle whether the geometric-series relation is quantitatively correct, and would clarify how much weight the analytical estimate should carry.","tokens_in":12580,"tokens_out":20050,"duration_ms":188178,"concrete_test":"Run additional GLE simulations at representative state points (U0=3 kBT, γM=0, τ/τ_D=1 and 10, τ_m/τ_D=0.001) and directly measure the velocity-direction commitors at x=0 by initiating trajectories with Maxwell-distributed velocities conditioned on sign. Compare the measured φ_X(0,v→X) with the geometric-series predictions 1/(2-κ) and (1-κ)/(2-κ) from SM V Eqs. (26)-(27), using κ from the Grote-Hynes formula Eq. (5) or from the measured reactive flux. Also compare the measured max[p(TP|x)] with Eq. (4). If these differ by more than the statistical error, the analytical estimate should be revised or removed, and the claim should rest on the direct simulation evidence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The analytical estimate Eq. (4) is derived in SM V from a geometric-series model in which each barrier recrossing attempt has an independent, constant success probability κ (SM V Eqs. 26-27). This ignores that for non-Markovian dynamics the outcome of an attempt is correlated with the velocity and history of previous attempts, and that κ as defined by Grote-Hynes is a flux-weighted rate ratio rather than a per-attempt probability. If the independence assumption fails, the relation max[p(TP|x)] = ((κ-1)^2+1)/(κ-2)^2 is not exact, and the quantitative connection between p(TP|x) and κ is unsupported. The paper itself reports deviations from Eq. (4) in the overdamped intermediate-memory regime. A related fragility is that the γM=0 limit requires m≠0 for well-posedness (SM IV); the simulations use τ_m/τ_D=0.001, so the long-memory approach to 1 is established only for this regularized finite-mass model, not for the strict overdamped non-Markovian limit. These gaps do not by themselves refute the simulation-based conclusion that max[p(TP|x)] can exceed 1/2, but they make the analytical support for the central claim conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the maximal value of the transition-path probability p(TP|x) for a one-dimensional reaction coordinate in a double-well potential under generalized Langevin dynamics with exponential memory friction and finite mass. It shows analytically and by simulation that max[p(TP|x)] is not bounded by the overdamped Markovian benchmark 1/2: it is nonmonotonic in the memory time, falls below 1/2 at intermediate memory in the overdamped regime, and approaches 1 for long memory or strong inertia. The authors propose an approximate relation, Eq. (4), connecting max[p(TP|x)] to the Grote-Hynes transmission coefficient kappa, test it against GLE simulations, and illustrate with the alpha3D protein that a peak near 1/2 does not imply Markovian dynamics. They conclude that p(TP|x) is not a reliable indicator of reaction-coordinate quality or Markovianity in non-Markovian systems.","tokens_in":12888,"tokens_out":9041,"duration_ms":85181,"significance":"If the conclusions hold, the paper corrects a common interpretation of a widely used metric: a p(TP|x) peak near 1/2 can occur even when the dynamics is strongly non-Markovian, as in the alpha3D example. The paper's strengths are the simple analytical estimate Eq. (4), the GH-based formulas, and the systematic GLE simulations covering inertial, Markovian, non-Markovian, and mixed-friction regimes. The derivations are compact, and the comparison with simulations uses the same model parameters with no fitted free parameters, which makes the qualitative message credible. The main weaknesses are the unproven independence assumption behind Eq. (4), missing statistical uncertainties in the simulation data, and some notation problems in Eq. (6); these affect the quantitative, but not necessarily the qualitative, conclusions.","major_comments":[{"comment":"The derivation of Eq. (4) assumes that each barrier recrossing attempt is an independent trial with the same success probability kappa, so that the velocity-direction commitor is kappa times sum over (1-kappa)^(2n). For non-Markovian dynamics this is an assumption, not a consequence of GH theory: kappa is a flux-weighted rate ratio, and the outcome of an attempt can be correlated with the velocity and with the history of earlier attempts through the memory kernel. The paper itself reports deviations from Eq. (4) in the overdamped intermediate-memory regime (Fig. 3B, tau_m/tau_D=0.001 near tau/tau_D ~ 1), so the formula should be framed as a heuristic estimate or supported by a direct test against simulation, e.g. by computing the velocity-direction commitor at the barrier and comparing it with kappa/(2-kappa).","section":"SM V, Eqs. (26)-(27)"},{"comment":"In Eq. (6), U''_max denotes the potential curvature at the barrier top, which is negative for the double-well potential; as written, kappa^2 = 1 - gamma/(tau U''_max) is larger than 1, and the associated square roots in SM VI (Eqs. (36)-(37)) are not well defined. If |U''_max| is intended, please define it explicitly and propagate the convention through Eq. (6) and SM VI. This is not merely cosmetic: Eq. (6) is the analytical basis for the tau -> infinity approach to 1 and for the tau* estimates in Fig. 3C.","section":"Eq. (6) and SM VI"},{"comment":"The simulation data are presented without error bars or confidence intervals. Since the central claims are quantitative (max[p(TP|x)] exceeds 1/2, drops below 1/2 at intermediate memory, and crosses 1/2 again at tau*), the absence of statistical uncertainties makes it difficult to assess whether the observed nonmonotonicity and the reported values of tau* are robust. Please add error bars or confidence intervals to all simulation markers, including Fig. 3C, using e.g. block averaging or bootstrapping.","section":"Figs. 2 and 3"},{"comment":"SM IV shows that the strict overdamped non-Markovian limit (m=0, gamma_M=0) is singular at the barrier, and the simulations therefore use tau_m/tau_D=0.001. Eq. (6) is an m -> 0 limit, but it does not by itself establish that the strict m=0 model has the same long-memory behavior. Please state explicitly whether the tau_m/tau_D -> 0 extrapolation at fixed tau/tau_D preserves max[p(TP|x)] -> 1 for tau -> infinity, or whether the finite-mass regularization is essential for the result; this would also clarify the comparison with the earlier overdamped result of Berezhkovskii and Makarov [35].","section":"SM IV and Figs. 2B/3"}],"minor_comments":[{"comment":"The statement that the results 'disqualif[y] p(TP|x) as a criterion for reaction coordinate quality' is too strong; the benchmark remains valid in the overdamped Markovian case. Suggest 'as a universal criterion' or 'as a criterion for systems with significant inertial or non-Markovian effects'.","section":"Abstract and Conclusion"},{"comment":"Typo: 'at there maximal value' should be 'at their maximal value'.","section":"Page 3, text near Fig. 2"},{"comment":"The notation U''_max is used both as the signed curvature at the barrier and as its magnitude; please state the sign convention explicitly and use |U''_max| where the square root or kappa^2 formula requires it.","section":"Eq. (6) and SM VI"},{"comment":"The analytical lines for Eq. (4) are described as colored dotted lines; please ensure each curve remains identifiable in grayscale printing, for example by combining line styles with distinct marker shapes in the legends.","section":"Fig. 2 and Fig. 3"},{"comment":"There appears to be an extra parenthesis in the expression <(p(TP|x=0,v))>_v; this is probably a typo for <p(TP|0,v)>_v.","section":"SM V, Eq. (28)"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely suitable for a letters-type journal in chemical physics, and the main correction to the literature is important. I would not require a full proof of Eq. (4) if it is explicitly reframed as a heuristic, because the simulations carry the qualitative claim; however, the notation error in Eq. (6), the missing statistical uncertainties, and the unstated role of the finite-mass regularization should be addressed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — here's my read on 2501.04512.\n\nThe central claim holds: for a GLE with exponential memory and finite mass, max[p(TP|x)] dips below 1/2 at intermediate memory and rises toward 1 at long memory, so the 1/2 benchmark is not a safe indicator of reaction-coordinate quality or Markovianity. The paper also reconciles this with Berezhkovskii-Makarov's earlier prediction of a strict decrease, and shows the difference comes from a small admixture of Markovian friction.\n\nWhat's genuinely new is the non-monotonic dependence on memory time and the analytical estimate Eq. (4) linking max[p(TP|x)] to the Grote-Hynes transmission coefficient. The limiting cases are correct: it recovers 1/2 for kappa=0 and 1 for kappa=1. The simulations are clean and cover the relevant parameter ranges, and the mixed-friction results in SM VII nicely explain why earlier work found max <= 1/2.\n\nSoft spots, in order of seriousness. First, the derivation of Eq. (4) in SM V assumes each recrossing attempt is independent with a constant success probability kappa. That's a geometric-series model, not a controlled expansion. For non-Markovian dynamics, velocity and history correlations could make the per-attempt probability vary, and Grote-Hynes kappa is a flux-weighted rate ratio, not obviously a per-attempt probability. The paper itself shows deviations from Eq. (4) in the overdamped intermediate-memory regime. So Eq. (4) should be treated as a physically motivated interpolation, not a derivation. Second, simulation data are shown without error bars; the trends are clear, but error bars would help, especially near the minima. Third, the pure non-Markovian limit gamma_M=0 requires finite mass for well-posedness (SM IV), so the long-memory approach to 1 is demonstrated for the regularized model with tau_m/tau_D=0.001, not for the strict overdamped non-Markovian limit. That's a real but minor caveat; the qualitative conclusion doesn't depend on it.\n\nThe citation pattern is honest: prior work [35] is engaged with directly, and the discrepancy is explained rather than glossed over. The protein example in Fig. 1 is illustrative and cited to Shaw group data, so no issue there.\n\nBottom line: this deserves a serious referee. It's a solid, useful paper for anyone using p(TP|x) as a quality metric. The heuristic nature of Eq. (4) and the missing error bars should be addressed in revision, but they don't undermine the main simulation-based conclusion.","headline":"Core claim holds: max[p(TP|x)] can exceed 1/2 and approach 1 for long memory in the finite-mass GLE, and the analytical estimate is useful though heuristic.","tokens_in":13388,"tokens_out":1710,"would_cite":true,"duration_ms":15794,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82C31"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that the standard 1/2 benchmark for transition-path probability fails once inertia or memory friction is present, with the true maximum approaching 1 for long memory.","keywords":["transition path probability","reaction coordinate quality","non-Markovian friction","generalized Langevin equation","memory effects","transmission coefficient","inertial dynamics","protein folding"],"falsifier":"In a GLE simulation with single-exponential memory, count the number of failed barrier-crossing attempts before a successful transition starting from the barrier top with a given initial velocity, and test whether this distribution is geometric with success probability $\\kappa$ from Grote-Hynes theory; a systematic mismatch would falsify the geometric-series assumption behind Eq. (4).","tokens_in":12402,"feed_emoji":"⚛️","tokens_out":5633,"duration_ms":53295,"temperature":0.7,"pith_summary":"This paper argues that the widely used benchmark value $\\max[p(\\mathrm{TP}|x)] = 1/2$, taken to indicate a good reaction coordinate in overdamped Markovian dynamics, is not reliable when inertia or non-Markovian memory friction is present. Using a generalized Langevin equation with a single-exponential memory kernel in a double-well potential, the authors show analytically and by simulation that the maximal transition-path probability is non-monotonic in the memory time: it dips below 1/2 at intermediate memory and approaches 1 when memory is long. The central analytical result connects this maximum to the transmission coefficient $\\kappa$ via the approximate formula $\\max[p(\\mathrm{TP}|x)] \\approx ((\\kappa-1)^2+1)/(\\kappa-2)^2$. If correct, this overturns a common assumption and implies that $p(\\mathrm{TP}|x)$ alone cannot certify reaction-coordinate quality or Markovianity.","feed_headline":"Memory friction pushes transition-path peaks above 1/2","feed_subtitle":"Theory and simulation show the 1/2 reaction-coordinate benchmark is crossed twice, so it cannot certify Markovianity.","key_machinery":"The central object is the conditional transition-path probability $p(\\mathrm{TP}|x)$, the fraction of equilibrium trajectories at position $x$ that belong to the transition-path ensemble; in the overdamped Markovian limit it equals $2\\phi_A(x)\\phi_B(x)$, giving a maximum of 1/2 at the barrier. The load-bearing identity is Eq. (4), $\\max[p(\\mathrm{TP}|x)] \\approx (\\kappa-1)^2+1)/(\\kappa-2)^2$, which relates the peak value to the transmission coefficient $\\kappa$ defined by Grote-Hynes rate theory as the ratio $k/k_\\mathrm{TST}$. The derivation treats each barrier-crossing attempt as an independent trial with success probability $\\kappa$, sums the geometric series of unsuccessful pairs of attempts, and thereby connects the velocity-direction committor at the barrier top to $\\kappa$. This formula, together with the Grote-Hynes result $\\kappa = \\lambda/\\omega_\\mathrm{max}$, is what carries the prediction that $\\max[p(\\mathrm{TP}|x)] \\to 1$ for long memory, $\\tau \\to \\infty$, in the overdamped non-Markovian limit.","core_discovery":"The paper's central claim is that $\\max[p(\\mathrm{TP}|x)]$, the peak value of the conditional transition-path probability along a reaction coordinate, can exceed the overdamped Markovian benchmark value of 1/2 and can approach 1 in the presence of either inertial dynamics or long non-Markovian memory friction. The authors derive an analytical estimate, Eq. (4), expressing this maximum in terms of the Grote-Hynes transmission coefficient $\\kappa$, which recovers 1 for $\\kappa = 1$ and 1/2 for $\\kappa = 0$. Their GLE simulations confirm that, for purely non-Markovian friction, the maximum decreases below 1/2 at intermediate memory times and then rises toward 1 as the memory time grows, so the value 1/2 is reached twice. They also show that adding even a small amount of explicit Markovian friction lowers the maximum, which reconciles their result with an earlier study that found $\\max[p(\\mathrm{TP}|x)] < 1/2$ in an overdamped non-Markovian setting. A protein-folding example illustrates the practical consequence: a peak near 1/2 can appear despite strongly non-Markovian dynamics.","pith_inferences":["If the central claim is right, reaction-coordinate validation protocols should either measure the memory time explicitly or use a criterion that accounts for inertia and memory, rather than relying on the absolute height of the $p(\\mathrm{TP}|x)$ peak.","The same geometric-series logic suggests that systems with multiple memory time scales could show more complex non-monotonicity, possibly with several crossings of the 1/2 line; this is a testable extension beyond the paper's single-exponential kernel.","One could invert Eq. (4) in practice: given a measured $\\max[p(\\mathrm{TP}|x)]$ and the barrier curvature, one can infer an effective transmission coefficient and, through the Grote-Hynes relation, an effective memory time from equilibrium trajectory data alone.","Single-molecule experiments on bistable systems with long correlation in the environmental fluctuations might observe $\\max[p(\\mathrm{TP}|x)] > 1/2$; a null result would constrain the effective memory time of the experimental coordinate."],"forward_implications":["The benchmark value 1/2 for $\\max[p(\\mathrm{TP}|x)]$ is crossed twice in non-Markovian dynamics, once at vanishing and once at long memory, so it cannot be used as a standalone indicator of reaction-coordinate quality or Markovianity.","In the purely non-Markovian overdamped limit, $\\max[p(\\mathrm{TP}|x)]$ approaches 1 as the memory time grows, independent of the inertial time scale.","Adding a small fraction of Markovian friction to non-Markovian friction lowers $\\max[p(\\mathrm{TP}|x)]$, which explains the difference from previous results that reported values below 1/2.","The analytical estimate based on the Grote-Hynes transmission coefficient provides a reference curve for $\\max[p(\\mathrm{TP}|x)]$ in inertial or non-Markovian systems, useful for interpreting simulation data.","For fast-folding proteins, where memory times can be comparable to folding times, a $p(\\mathrm{TP}|x)$ peak near 1/2 does not imply Markovian dynamics."],"supporting_citations":[{"why":"Introduced $p(\\mathrm{TP}|x)$ as a reaction-coordinate quality measure and derived the overdamped Markovian relation $p(\\mathrm{TP}|x)=2\\phi_A\\phi_B$ that sets the 1/2 benchmark.","marker":"[15]"},{"why":"Earlier single-molecule Markovianity test that claimed non-Markovian dynamics keeps $\\max[p(\\mathrm{TP}|x)]$ below 1/2; the paper reconciles its contrary finding with this result.","marker":"[35]"},{"why":"Provides the Grote-Hynes transmission coefficient $\\kappa$ and reactive frequency used in the central analytical estimate, Eq. (4).","marker":"[36]"},{"why":"Supplies the $\\alpha_3$D protein folding data and the non-Markovian model showing a memory time comparable to the folding time, used to demonstrate the spurious 1/2 peak.","marker":"[30]"},{"why":"Establishes non-Markovian modeling of protein folding and memory-induced effects, providing the biophysical context for the paper's claims.","marker":"[28]"},{"why":"Documents memory-induced acceleration and slowdown of barrier crossing, supplying the inertial and memory time scales used in the analysis.","marker":"[40]"},{"why":"Numerical study of barrier-crossing times for non-Markovian friction showing Grote-Hynes theory accurately describes dynamics on the barrier.","marker":"[43]"},{"why":"Projection-operator derivation of the generalized Langevin equation underpins the model used for the simulations.","marker":"[33]"}],"fun_headline_variants":["Non-Markovian memory lifts transition-path peak above 1/2","Memory friction makes transition-path probability exceed 1/2","Reaction-coordinate benchmark 1/2 fails under non-Markovian memory","Transition-path peak crosses 1/2 twice with long memory time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analytical estimate assumes that every failed barrier-crossing attempt is independent and has the same chance of succeeding, so the number of attempts follows a simple geometric series; if attempts are correlated or the success chance changes with velocity or history, the predicted maximum $p(\\mathrm{TP}|x)$ could be off, even though the qualitative simulation trend may survive.","fun_headline_variants_meta":{"raw":{"variants":["Non-Markovian memory lifts transition-path peak above 1/2","Memory friction makes transition-path probability exceed 1/2","Reaction-coordinate benchmark 1/2 fails under non-Markovian memory","Transition-path peak crosses 1/2 twice with long memory time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000694,"raw_usage":{"total_tokens":3120,"prompt_tokens":906,"completion_tokens":2214,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":522,"completion_tokens_details":{"reasoning_tokens":2135}},"tokens_in":522,"tokens_out":2214,"duration_ms":15197,"temperature":1.0,"reasoning_tokens":2135,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:30:11.383018+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a GLE simulation with single-exponential memory, count the number of failed barrier-crossing attempts before a successful transition starting from the barrier top with a given initial velocity, and test whether this distribution is geometric with success probability $\\kappa$ from Grote-Hynes theory; a systematic mismatch would falsify the geometric-series assumption behind Eq. (4).","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduced $p(\\mathrm{TP}|x)$ as a reaction-coordinate quality measure and derived the overdamped Markovian relation $p(\\mathrm{TP}|x)=2\\phi_A\\phi_B$ that sets the 1/2 benchmark."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier single-molecule Markovianity test that claimed non-Markovian dynamics keeps $\\max[p(\\mathrm{TP}|x)]$ below 1/2; the paper reconciles its contrary finding with this result."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Grote-Hynes transmission coefficient $\\kappa$ and reactive frequency used in the central analytical estimate, Eq. (4)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the $\\alpha_3$D protein folding data and the non-Markovian model showing a memory time comparable to the folding time, used to demonstrate the spurious 1/2 peak."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes non-Markovian modeling of protein folding and memory-induced effects, providing the biophysical context for the paper's claims."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents memory-induced acceleration and slowdown of barrier crossing, supplying the inertial and memory time scales used in the analysis."},{"cited_title":"Zwanzig, Ensemble Method in the Theory of Irre- versibility, J","cited_arxiv_id":null,"evidence_quote":"Numerical study of barrier-crossing times for non-Markovian friction showing Grote-Hynes theory accurately describes dynamics on the barrier."},{"cited_title":"Memory and Friction: From the Nanoscale to the Macroscale","cited_arxiv_id":"2410.22588","evidence_quote":"Projection-operator derivation of the generalized Langevin equation underpins the model used for the simulations."}],"review_version":1}