{"id":"66d9aee3-fdd5-4d4d-bba5-2d475c189f27","arxiv_id":"2506.05966","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A 2D generalized Langevin equation with matrix memory predicts coupled dihedral kinetics in pentane better than 1D models, mostly through the 2D potential, while alanine dipeptide rates gain nothing from the 2D treatment.","lead":"This paper extends the memory-dependent generalized Langevin equation to two coupled reaction coordinates and tests it against explicit-water simulations of pentane and alanine dipeptide. It finds that the two-dimensional energy landscape, more than the off-diagonal friction, drives kinetic corrections in pentane, while one-dimensional models suffice for alanine dipeptide rates.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The pentane MFPT attribution to the 2D potential rests on Eq. (8)'s Gaussian FDR, which is known to fail for non-linear observables; App. O shows such a failure nearby, so the ablation could partly be a noise-model artifact. A direct FDR/non-Gaussianity test would settle this.","rationale":"Good-faith reading: the paper is a careful computational study within an established GLE program; it openly flags the FDR approximation and asks for future non-Gaussian treatments. The central claim is supported by multiple observables (MFPT, MSD, cross-displacement, transition-path times) and by a clean internal ablation. The weakest premise is indeed Eq. (8), as the reader notes; the paper's own App. O shows the Gaussian ansatz can fail for a related 1D coordinate, which makes the concern concrete rather than hypothetical. However, the ablation and the transition-path-time analysis compare simulations that share the same noise ansatz, so the relative ranking of the 2D-potential vs off-diagonal-friction terms is less sensitive to the absolute FDR error than a single MFPT comparison. The paper also explicitly moderates its claim ('accurate yet not perfect') and identifies the approximation as the cause of residual deviations. I therefore do not see grounds to overturn the reader's ACCEPT; the right outcome is to keep the verdict and to note the direct FDR/non-Gaussianity check as a robustness condition for the strongest interpretation of the claim.","tokens_in":30357,"tokens_out":15752,"duration_ms":158894,"concrete_test":"From the pentane MD trajectory, compute the residual random force F_R(t) = M \\ddot x(t) + \\nabla U(x(t)) + \\int_0^t Gamma(t-s) \\dot x(s) ds using the extracted kernel, and test (i) whether <F_R(t) F_R^T(0)> = k_BT Gamma(t) (Eq. 8) and (ii) whether the noise is Gaussian (e.g., excess kurtosis of each component). Then rerun the Fig. 3(d) ablation with a non-Gaussian noise model matched to the measured F_R statistics (e.g., Ref. [69]). If the ranking '2D potential dominates MFPT' survives, the concern is resolved; if not, the central claim must be qualified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim that for pentane the multi-dimensional coupled potential, not off-diagonal friction, is the key factor in reproducing MD MFPTs is established by the ablation in Fig. 3(d): zeroing Gamma12 and M12 while keeping U(theta1, theta2) (orange) matches the full 2D GLE (green), whereas decoupling the potential (blue) does not. This ablation is performed entirely inside Markovian embedding simulations whose noise is Gaussian with covariance kBT Gamma(t) (Eq. 8). Section 3.2 explicitly notes that Eq. (8) 'is generally not fulfilled for non-linear observables,' and the Discussion adds that the Gaussian ansatz is 'approximate,' citing App. O for a 1D coordinate (x_avg) where Gaussian FDR fails badly. If the FDR error is configuration-dependent, which is plausible for non-linear dihedral observables, then the 2D potential changes which configurations are sampled (it forbids the cis-cis states), so the two legs of the ablation are subject to different noise-model errors. The observed slowdown in orange could then be partly an artifact of the Gaussian ansatz being less wrong in the allowed region, rather than a genuine property of the 2D confinement. The cross-displacement and transition-path-time evidence (Fig. 3c, App. N) reduce, but do not eliminate, this possibility.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a multi-dimensional generalized Langevin equation (GLE) framework with a matrix memory kernel and a Markovian embedding scheme, and applies it to the two-dimensional dihedral dynamics of solvated pentane and alanine dipeptide. The authors extract the memory matrix from MD trajectories via a Volterra equation, fit it to a sum of matrix exponentials, and simulate the resulting GLE with off-diagonal friction and mass coupling. For pentane, they find that a full 2D GLE reproduces MD mean first-passage times (MFPTs) substantially better than a 1D GLE, and an ablation (Fig. 3d) indicates that the dominant factor is the multi-dimensional coupled potential rather than the off-diagonal friction or mass. For alanine dipeptide, the 2D GLE provides no MFPT advantage over separate 1D GLEs, although off-diagonal friction is still needed to reproduce cross-displacement correlations. The paper includes extensive validation: a model-system test of the extraction (Appendix B), tests of position-dependent mass (Appendix F), exponential-oscillatory memory fits (Appendix K), alternative 1D reaction coordinates (Appendix O), and a frozen-carbon freeze analysis (Appendix P).","tokens_in":25,"tokens_out":10019,"duration_ms":119588,"significance":"If the central claim holds, the paper makes a useful contribution by delineating when multi-dimensional memory effects are kinetically relevant: for pentane, the off-diagonal friction is significant in magnitude but is not the main cause of the 2D-over-1D MFPT improvement, whereas the coupled potential is. For alanine dipeptide, the 2D coupling is irrelevant for individual MFPTs but essential for cross-correlations. The paper is careful and transparent: it ships a validated extraction scheme (Appendix B), tests several modeling choices (Appendices F, K, O, P), and compares out-of-sample MFPT predictions against MD. The main limitation, acknowledged by the authors, is that the Markovian embedding uses a Gaussian random force with the approximate fluctuation–dissipation relation of Eq. (8), which is generally not exact for non-linear observables. This limits but does not remove the value of the numerical conclusions, provided the Gaussian ansatz is adequate for the specific observables studied; the paper does not test this directly.","major_comments":[{"comment":"The central attribution for pentane—that the multi-dimensional coupled potential, not off-diagonal friction, is the key factor reproducing MD MFPTs—rests entirely on Markovian embedding simulations whose noise is Gaussian with covariance given by the approximate fluctuation–dissipation relation in Eq. (8). The paper itself notes in Section 3.2 that Eq. (8) 'is generally not fulfilled for non-linear observables' and Appendix O demonstrates a concrete failure for the averaged dihedral coordinate x_avg. Because the 2D potential changes which configurations are sampled (it eliminates the cis-cis states), the orange and blue legs of the ablation in Fig. 3(d) are subject to potentially different noise-model errors. The observed slowdown in the orange curve could therefore be partly an artifact of the Gaussian ansatz being less inappropriate in the allowed region, rather than a genuine kinetic effect of the 2D confinement. I request a direct test of the validity of Eq. (8) for the 2D dihedral observables, for example by comparing the MD-derived random-force distribution with the Gaussian prediction, or by repeating the ablation with a non-Gaussian noise model from Ref. [69]. Without such a test, the quantitative MFPT agreement and the ranking in Fig. 3(d) could reflect error cancellation rather than the claimed physics.","section":"Section 3.2, Eq. (8), Fig. 3(d)"}],"minor_comments":[{"comment":"The phrase 'Unlike previous studies, our results highlight the critical role of different terms in the multi-dimensional GLE' is vague; it would be clearer to specify which terms (coupled potential versus off-diagonal friction) matter for which observables.","section":"Abstract"},{"comment":"The text in Section 3.2 states that the off-diagonal mass entry is 'a factor of around -0.05' of the diagonal entry, but the values in Eq. (17) give M12/M11 = -9.69e-7 / 2.56e-5 ≈ -0.038. Please correct the wording or clarify the rounding.","section":"Section 3.1, Eq. (17)"},{"comment":"The notation in Eq. (C6) is unclear: the left side ⟨y_kl_i(0) y_kn_i(0)⟩ uses three indices on y, and the right side is a double Kronecker delta with indices k,l,n, which seems inconsistent with the preceding definitions. Please rewrite this equation in terms of vector components, e.g., ⟨y_{i,k}(0) y_{i,l}(0)⟩.","section":"Appendix C, Eq. (C6)"},{"comment":"The off-diagonal memory times τ12_i are all reported as 0.001 ps, the smallest value in the tables. Please state whether this is a fitted value, a lower bound imposed by the optimizer, or a fixed parameter, and if fixed, whether the results are sensitive to that choice.","section":"Table 1 and Table 2"},{"comment":"The symbols '□k⁄=l' in the figure caption are rendered incorrectly; they should be Γ_{k≠l} (and M_{k≠l}). This appears to be a font/rendering issue, but the caption should be legible in the final version.","section":"Figure 3 caption"},{"comment":"The statement that code and input files are 'available from the corresponding author upon request' is less reproducible than depositing them in a public repository (e.g., Zenodo or GitHub). Given the paper's emphasis on validation, a permanent DOI would be preferable.","section":"Data Availability Statement"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically strong and the authors are clearly experts in this area. The central claim is well-supported within the Gaussian-noise modeling framework, and the extensive appendices are a major strength. My main concern is that the load-bearing assumption of Eq. (8) is not tested for the 2D dihedral observables, and the paper itself shows a nearby counterexample (Appendix O). I believe this is fixable within the manuscript's scope by adding a direct FDR/non-Gaussianity check or a sensitivity study with a non-Gaussian noise model. I would not recommend rejection; the manuscript is close to publishable, but the requested check is important for the confidence in the central attribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look. The paper's real contribution is the explicit 1D-vs-2D GLE comparison and the ablation separating the 2D potential from off-diagonal friction. For pentane, the full 2D GLE reproduces MD first-passage times much better than a 1D GLE, and the ablation says it is the coupled potential, not the off-diagonal memory, that does the work. For alanine dipeptide, the 2D GLE is no better than 1D for rates, but off-diagonal friction is needed for cross-displacements. Those are clean, useful results.\n\nThe method section is solid: the memory-matrix extraction is a direct Volterra inversion with a verification on a model system with a known kernel (App B), and the paper runs several sensible checks—position-dependent mass, exponential-oscillatory fits, a frozen-carbon control. The MFPT comparisons are out-of-sample in the sense that memory parameters come from correlation functions, not from the target kinetics. The paper is also honest about its main approximation: Eq. (8), the Gaussian fluctuation-dissipation relation, is generally not fulfilled for non-linear observables, and the Discussion says the random force is approximate.\n\nThe soft spot is exactly that premise. The ablation that pins the pentane slowdown on the 2D potential is performed inside a Markovian embedding whose noise is Gaussian with covariance kBT times the fitted kernel. App O shows a nearby 1D coordinate (the averaged dihedral) where the Gaussian ansatz fails badly. If the FDR error is configuration-dependent, the orange and blue legs of the ablation could in principle be subject to different noise-model errors, and the slowdown attributed to the 2D potential could be partly an artifact. That is a legitimate caveat, but I do not think it overturns the conclusion: the transition-path-time distributions in App N independently support the 2D-potential effect, and the full 2D GLE matches the MD cross-displacements. The quantitative claim should be read as conditional on the Gaussian embedding, which the paper itself says.\n\nMinor issues: code and data are 'upon request' only, and error bars are not visible in most figures (the text says they are smaller than linewidth). Neither is serious.\n\nThis is a careful, honest paper. The right audience is the coarse-grained dynamics and GLE community; they will get a clear map of when multi-dimensionality matters. It deserves a serious referee. My recommendation: send it to review, and ask the authors to run a direct test of the FDR/non-Gaussianity of the two dihedral observables, to close the one real gap.","headline":"Clean ablation shows the 2D potential, not off-diagonal friction, drives pentane MFPT improvement—but the Gaussian noise ansatz is the load-bearing premise.","tokens_in":31325,"tokens_out":3825,"would_cite":true,"duration_ms":38116,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"For pentane's two coupled dihedrals, a 2D generalized Langevin equation beats 1D models mainly because the 2D potential confines both angles, not because of off-diagonal friction.","keywords":["generalized Langevin equation","multi-dimensional reaction coordinates","non-Markovian memory","friction coupling matrix","dihedral isomerization dynamics","mean first-passage time","Markovian embedding","coarse-grained dynamics"],"falsifier":"Run the 2D GLE simulation of pentane with the fitted matrix memory and mass but with the potential replaced by $U_1(\\theta_1)+U_2(\\theta_2)$: the paper predicts the MFPTs should remain close to the 1D-GLE values, so agreement with MD in that run would refute the claim that the coupled potential is the key factor.","tokens_in":30132,"feed_emoji":"🔬","tokens_out":7106,"duration_ms":65804,"temperature":0.7,"pith_summary":"This paper asks when a coarse-grained model must keep two reaction coordinates together rather than treating each coordinate with its own one-dimensional generalized Langevin equation. For pentane in water, a 2D GLE with a matrix memory kernel reproduces the molecular-dynamics mean first-passage times much better than 1D GLEs, and the paper shows the improvement comes primarily from the coupled 2D potential, which sterically forbids combining the two cis states. Off-diagonal friction is real and negative, but its main visible effect is to reproduce cross-correlations between the dihedral angles, not the barrier-crossing rates. For alanine dipeptide, whose 2D potential is nearly separable, the 2D GLE offers no rate improvement over 1D GLEs. The methodological result is a tractable recipe: extract the full memory matrix from trajectories, fit it by matrix exponentials, and simulate the Markovian embedding.","feed_headline":"2D potential, not friction, drives pentane GLE kinetics","feed_subtitle":"For pentane's two dihedrals, a coupled 2D potential, not off-diagonal friction, makes the 2D GLE beat 1D models.","key_machinery":"The central object is the memory kernel matrix $\\hat{\\Gamma}(t)$ in the multi-dimensional GLE, whose off-diagonal entries couple the frictional forces on different reaction coordinates. The paper extracts it from MD trajectories by iteratively solving a Volterra equation for the running integral $\\hat{G}(t)$, fits it as a sum of matrix exponentials, and simulates the equivalent Markovian embedding. The decisive comparison isolates the influence of the multi-dimensional potential $U(\\theta_1,\\theta_2)$ and the off-diagonal entries $\\Gamma_{12}$ and $M_{12}$ by switching each off separately in the simulations.","core_discovery":"The paper's central claim is that for the two coupled dihedral angles of pentane in water, the full two-dimensional GLE with a matrix memory kernel reproduces the MD mean first-passage times significantly better than uncoupled 1D GLE models, and that this improvement is dominated by the coupled 2D potential of mean force rather than by the off-diagonal friction entries. The off-diagonal friction is nonetheless substantial, about 18% of the diagonal value and negative for pentane, and it is necessary to reproduce cross-correlations between the two dihedral coordinates. For alanine dipeptide, the off-diagonal friction is positive and roughly a third of the diagonal value, yet the 2D GLE gives no MFPT improvement over 1D GLEs because its 2D potential is nearly separable. The paper also shows that freezing pentane's inner carbons removes the off-diagonal friction, indicating an intramolecular origin for that coupling.","pith_inferences":["Editorial extension: the sign and magnitude of the off-diagonal friction matrix could serve as a mechanistic fingerprint, with negative coupling tied to intramolecular motion (it vanishes when the inner carbons are frozen) and positive coupling in alanine dipeptide attributed to solvent-mediated hydrodynamic interactions; varying solvent viscosity should alter mainly the positive component.","Editorial extension: the paper's decomposition implies a cheap screening test for whether a multi-dimensional GLE is needed for rates: compute the difference between the 2D potential of mean force and the sum of the 1D potentials; if the difference is concentrated at sterically forbidden states, rates require the 2D potential, whereas strong off-diagonal friction alone will not.","Editorial extension: the failure of the averaged 1D coordinate suggests that dimensionality reduction can push non-Gaussianity into the random force; a testable follow-up is to measure the orthogonal-force non-Gaussianity for such 1D projections and check whether it predicts the MFPT error seen in the paper."],"forward_implications":["For pentane, replacing the 2D potential by the decoupled sum $U_1+U_2$ removes most of the 2D-over-1D MFPT improvement, so coupled potential confinement is the controlling term for this molecule.","Off-diagonal friction and mass must be retained to reproduce the mean cross-displacement between the two coordinates; a diagonal-only 2D GLE fails at short times.","For near-separable 2D potentials such as alanine dipeptide, independent 1D GLEs are sufficient for MFPTs, and a 2D GLE is not advantageous for rates.","Multi-exponential matrix fits are sufficient for MFPT prediction; adding exponential-oscillatory terms to the memory kernel does not improve the rates.","Freezing the inner carbons of pentane makes off-diagonal friction negligible, indicating that the coupling arises from intramolecular motion rather than hydrodynamics."],"supporting_citations":[{"why":"Supply the hybrid-projection derivation of the multi-dimensional GLE and the approximate fluctuation-dissipation relation used in equation (8).","marker":"[28, 29]"},{"why":"Introduced the multi-dimensional GLE with position-dependent mass and the Markovian embedding that this paper simplifies and compares against 1D GLEs.","marker":"[57]"},{"why":"Provided the 1D Volterra iteration and Markovian embedding that the paper generalizes to matrices and uses for its 1D baselines.","marker":"[18]"},{"why":"Showed butane dihedral dynamics are dominated by internal friction; used to interpret memory-kernel oscillations and the frozen-carbon experiment.","marker":"[52]"},{"why":"Established memory speed-up regimes and multi-exponential embedding, used to explain the Langevin slowdown and justify the fitting ansatz.","marker":"[67, 68]"},{"why":"Defined MFPT and transition-path-time measurement with recrossings and memory-dependent friction, used for the kinetic comparisons.","marker":"[23]"},{"why":"Brownian dynamics of n-alkanes with correlated neighboring dihedral transitions, cited for the intramolecular origin of off-diagonal friction.","marker":"[36]"}],"fun_headline_variants":["2D potential, not friction, drives pentane GLE accuracy","Coupled potential, not friction, key to pentane 2D GLE","Pentane MFPTs: 2D potential wins over off-diagonal friction","For pentane, 2D GLE gains come from potential, not friction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole simulation protocol hinges on assuming that a Gaussian random force with variance $k_BT$ times the fitted memory kernel reproduces the actual random force acting on these nonlinear dihedral coordinates, and the paper notes this relation generally fails for nonlinear observables.","fun_headline_variants_meta":{"raw":{"variants":["2D potential, not friction, drives pentane GLE accuracy","Coupled potential, not friction, key to pentane 2D GLE","Pentane MFPTs: 2D potential wins over off-diagonal friction","For pentane, 2D GLE gains come from potential, not friction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000577,"raw_usage":{"total_tokens":2730,"prompt_tokens":964,"completion_tokens":1766,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":580,"completion_tokens_details":{"reasoning_tokens":1682}},"tokens_in":580,"tokens_out":1766,"duration_ms":12418,"temperature":1.0,"reasoning_tokens":1682,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:14:01.844910+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the 2D GLE simulation of pentane with the fitted matrix memory and mass but with the potential replaced by $U_1(\\theta_1)+U_2(\\theta_2)$: the paper predicts the MFPTs should remain close to the 1D-GLE values, so agreement with MD in that run would refute the claim that the coupled potential is the key factor.","supporting_citations":[{"cited_title":"Hegger and G","cited_arxiv_id":null,"evidence_quote":"Introduced the multi-dimensional GLE with position-dependent mass and the Markovian embedding that this paper simplifies and compares against 1D GLEs."},{"cited_title":"Reichert and H","cited_arxiv_id":null,"evidence_quote":"Showed butane dihedral dynamics are dominated by internal friction; used to interpret memory-kernel oscillations and the frozen-carbon experiment."},{"cited_title":"Klippenstein, M","cited_arxiv_id":null,"evidence_quote":"Defined MFPT and transition-path-time measurement with recrossings and memory-dependent friction, used for the kinetic comparisons."},{"cited_title":"Abrash, S","cited_arxiv_id":null,"evidence_quote":"Brownian dynamics of n-alkanes with correlated neighboring dihedral transitions, cited for the intramolecular origin of off-diagonal friction."}],"review_version":1}