REVIEW 3 major objections 5 minor 3 references
Investigation on non-ergodicity of protein dynamics
T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash
Pith's one-line read Using 100 replicas of two proteins, this paper argues that protein dynamics is completely ergodic in the picosecond-to-nanosecond window, and that reported non-ergodicity reflects incomplete convergence.
desk verdict Useful finite-time ergodicity check with overreach in the headline claim; EB decrease alone cannot prove convergence to zero. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central objects are the ergodicity breaking parameter $EB(T,t)=\mathrm{Var}[\delta_k^2(T,t)]/\langle\delta_k^2(T,t)\rangle^2$ and the distribution $\phi(\xi)$ of the replica-normalized time-averaged mean square displacement $\xi_k(T,t)=\delta_k^2(T,t)/\langle\delta^2(T,t)\rangle$. The argument runs by showing that $EB$ decreases monotonically toward zero as $T/t$ grows (with $T=100$ ns and $t\le 1$ ns), and that $\phi$ narrows to a Gaussian centered at 1 without weight at zero; the paper takes these as the standard signatures of ergodicity, and uses the collapse of ensemble-averaged MSD onto the ensemble-averaged TA-MSD as a direct check.
What would settle it
Compute the ergodicity-breaking parameter for the same trajectories at lag times approaching the trajectory length, for instance T/t = 2 and 5; if EB stops decreasing and levels off at a nonzero value, the conclusion that the dynamics is completely ergodic in this window would be wrong.
Extended reading notes
Core claim
The paper's central claim is that, over timescales from picoseconds to nanoseconds, a folded protein explores its available native-state conformations in a fully ergodic manner, and that apparent non-ergodicity reported by earlier work (notably Li et al., ref 24) is an artifact of incomplete convergence of the measured quantities. The evidence is drawn from 100 independent 100-ns replica trajectories of villin headpiece (35 residues) and the CAP-cAMP homodimer (~400 residues). For each replica the paper computes a time-averaged mean square displacement (TA-MSD) and compares it with the ensemble average over replicas. The ergodicity breaking parameter $EB$ — the variance of TA-MSD divided by its squared ensemble average — falls monotonically below $10^{-2}$ as $T/t$ grows, and the distribution of the ratio $\xi = \delta_k^2 / \langle \delta^2 \rangle$ is a narrow Gaussian centered at 1 with no weight at zero. The paper concludes these are the signatures of an ergodic system and that the size dependence suggested by Li et al. does not appear in this time window.
Load-bearing premise
The argument hinges on reading the monotonic decrease of the ergodicity-breaking parameter below $10^{-2}$, within trajectories of 100 ns and lag times no larger than 1 ns, as evidence that it would converge to zero rather than to a small nonzero plateau outside the sampled range.
Editorial extensions
If this is right
- Simulation time averages and experimental ensemble averages can be compared directly for folded proteins within the pico-to-nanosecond window.
- The size-dependent non-ergodicity reported by Li et al. does not hold for these two systems in this window; the apparent effect likely comes from incomplete statistical convergence.
- The multi-funnel interpretation of static heterogeneity implies that ergodicity within a native-state valley is compatible with non-ergodicity between valleys.
- The low EB values justify using ensemble-averaged TA-MSD rather than single-trajectory TA-MSD in analyses of subdiffusion in this timescale.
- The conclusion is limited to the sampled native states; potential non-ergodicity over larger timescales remains open.
Reading between the lines
- If the ergodicity claim holds for these two proteins, it likely extends to other globular proteins in the same timescale, but only for descriptors that equilibrate quickly; slow collective variables may still show replica-to-replica spread.
- A sharper test of the paper's conclusion would be to compute EB for the same trajectories using lag times approaching T (e.g., T/t = 2), where a non-ergodic system would reveal a plateau.
- The multi-funnel picture proposed in the discussion predicts a bimodal distribution of escape times — fast intrabasin and slow interbasin — which could be tested by long single-molecule trajectories or by enhanced-sampling simulations.
- The paper's reliance on a single MD protocol leaves open the question of force-field sensitivity; rerunning the analysis with different force fields would clarify whether ergodicity in this window is a generic property or protocol-dependent.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper uses 100 independent 100-ns all-atom MD simulations of two proteins, the 35-residue Villin headpiece and the ~400-residue CAP homodimer, to test whether protein dynamics is ergodic in the picosecond-to-nanosecond window. It computes the ensemble-averaged mean-square displacement and the time-averaged MSD of individual replicas, the ergodicity-breaking parameter EB (Eq. 6), and the distribution phi of normalized time-averaged MSDs. The authors report that EB decreases below 10^-2 with increasing T/t and that phi is a narrow distribution centered at 1, and they conclude that protein dynamics is "completely ergodic" in this window irrespective of protein size, with deviations from ergodicity attributed to incomplete convergence. The Discussion, however, concedes that if multiple native valleys exist, a single protein molecule is "by definition non-ergodic" because it cannot visit another valley, a statement that conflicts with the abstract's absolute wording.
Significance. If established, the result would challenge the interpretation of recent single-molecule experiments (Li et al., Ye et al.) as evidence of non-ergodicity in the short-time regime and would support the common practice of comparing short MD trajectories with ensemble-averaged experimental observables. The paper's strengths are that EB and phi are computed directly from trajectories using standard definitions with no fitted parameters entering the central comparison, and the two-protein design with very different sizes directly addresses the previously proposed size dependence. The main weakness is that the finite-time data cannot distinguish ergodic convergence to zero from a transient approach to a nonzero plateau, and the headline claim of complete ergodicity is internally contradicted by the paper's own multi-valley discussion.
major comments (3)
- [Discussion (final paragraph) and Abstract] The abstract and conclusion state that protein exhibits a "completely ergodic dynamic within a time window ranging from pico-to-nanoseconds, irrespective of their size," but the Discussion explicitly says that if multiple native valleys exist, a single protein molecule is "by definition non-ergodic" because it cannot change its native state by visiting another valley. Since the simulations reported here are, on the paper's own model, confined to a single valley, they cannot support the word "completely." The defensible conclusion is that the dynamics is locally ergodic within the sampled valley and time window, not globally ergodic; the overclaim should be removed or substantially qualified.
- [Methods and Results, Fig. 3 and Eqs. 6-7] The central evidence for ergodicity is the monotonic decrease of EB with T/t, with T up to 100 ns and lag times t up to 1 ns, so T/t is at most 100. This finite-time trend is equally consistent with an ergodic system converging to zero and with a non-ergodic system whose EB would eventually plateau at a nonzero value because trajectories remain trapped in one valley on the sampled timescale. The paper does not demonstrate that the EB plateau is zero; it only shows that EB is below 10^-2 over the sampled range. Therefore the claim that "deviations from ergodic behavior are due to incomplete convergence" is not established by the presented data. A quantitative test, such as an extrapolation of EB to the T/t -> infinity limit, a fit to a known functional form, or a comparison with a non-ergodic model over the same range, is needed.
- [Methods, first paragraph] The entire MD simulation protocol is deferred to reference 12. The manuscript does not state the force field, simulation package, water model, thermostating/barostating scheme, equilibration procedure, or how the 100 independent replica starting configurations were extracted from the 1-microsecond simulation. Because the central conclusion depends on these trajectories adequately sampling the native-state dynamics of both proteins, the results are not reproducible from the manuscript as written, and the applicability of the previous protocol to these two systems cannot be assessed. At minimum, a summary of the protocol and a statement of replica-generation details should be included.
minor comments (5)
- [Throughout] The text contains numerous ligature and OCR artifacts, such as "signi5cant," "5ndings," and "simualted," which should be corrected in a polished manuscript.
- [Figure 1 caption] The caption contains typographical errors, including "segement" and "beetween," and the phrase "distance beetween the arginines" should be reworded for clarity.
- [Methods and Results, paragraph after Eq. 6] The statement that EB below 10^-2 means the variance of the TA-MSD is "at least one tenth" of the squared average is backwards: from Eq. 6, EB < 10^-2 implies Var < 10^-2 <delta^2>^2, i.e., the variance is less than one hundredth of the squared average, which would actually strengthen the authors' qualitative claim.
- [Fig. 3] No error bars or confidence intervals are reported for the EB estimates; with 100 replicas, the sampling uncertainty of EB should be quantified so that the reader can judge whether the decrease with T/t is significant.
- [Fig. 5 and surrounding text] The normalized root-mean-square errors of 15% and 11% are reported without any null model or confidence interval, so the reader cannot assess whether these values are small enough to support the claim of equality between the two MSD estimates.
Circularity Check
No circular reasoning: the ergodicity claim is an empirical inference from directly computed statistical quantities, with no fitted input renamed as prediction.
full rationale
The paper's central quantities—the ergodicity-breaking parameter EB (Eqs. 6–7) and the distribution phi of xi—are computed directly from the 100 independent MD trajectories using standard definitions (Eqs. 2 and 3). The ergodicity conclusion follows from observing that EB decreases with T/t and that phi narrows around 1. These are empirical observations, not outputs of a model parameter fitted to the conclusion. The only self-citation is to the authors' previous work (ref. 12) for simulation details and for the T/t > 100 rule of thumb; neither assumes the target result that protein dynamics is ergodic on the pico-to-nanosecond timescale. No equation in the paper reduces to its own input by construction, and no fitted parameter is renamed as a prediction. The paper's limitation—that a monotonic decrease of EB over a finite window may not distinguish ergodic convergence from a non-ergodic plateau beyond the sampled range—is a scientific validity concern about extrapolation, not a circularity in the derivation. Likewise, the Discussion's concession that a protein in a multi-funnel landscape is non-ergodic by definition is an internal inconsistency with the abstract's wording, but it does not make the pico-to-nanosecond ergodicity claim circular. Therefore, no circular step is present.
Assumptions & free parameters
free parameters (2)
- T/t ratio threshold =
>= 100
- Number of replicas N =
100
assumptions (4)
- domain assumption The ergodicity-breaking parameter and phi distribution diagnose ergodicity
- domain assumption The 100 ns replicas sample the relevant equilibrium dynamics
- ad hoc to paper The deferred MD protocol in ref. 12 is valid and applicable
- ad hoc to paper The sampled T/t range reaches the ergodic asymptotic regime
Cite this review
Pith. "Pith review of Investigation on non-ergodicity of protein dynamics." pith.science (2026). https://pith.science/paper/7S7HAJK7
@misc{pith2026250108754,
author = {Pith},
title = {Pith review of: Investigation on non-ergodicity of protein dynamics},
year = {2026},
howpublished = {\url{https://pith.science/paper/7S7HAJK7}},
note = {Machine review of arXiv:2501.08754}
}
read the original abstract
The study of microscopic protein dynamics has historically presented significant challenges to researchers seeking to develop a comprehensive and detailed description of its diverse and intriguing features. Recent experimental and theoretical studies have proposed the hypothesis that protein dynamics may be non-ergodic. The implications of this finding are of paramount importance from both a practical and theoretical standpoint. In this study, we employ all-atom molecular dynamics simulations to examine these results over a time window spanning from picoseconds to nanoseconds. To this end, we utilize widely used statistical tools. Our findings challenge the conclusions of previous studies, which suggested that proteins exhibit non-ergodic dynamics. Instead, we demonstrate that deviations from ergodic behavior are due to incomplete convergence of the investigated quantities. Additionally, we discuss the implications of findings that suggest a potential breaking of the ergodic hypothesis over larger time windows, which were not directly investigated in this study.
Reference graph
Works this paper leans on
-
[2]
Frauenfelder, H., Sligar, S. G. & Wolynes, P . G. The Energy Landscapes and Motions of Proteins. Science (1979) 254, 1598–1603 (1991). 3. Iben, I. E. T. et al. Glassy behavior of a protein. Phys Rev Lett 62, 1916–1919 (1989). 4. Mori, T. et al. Detection of boson peak and fractal dynamics of disordered systems using terahertz spectroscopy. Phys Rev E 102,...
work page 1979
-
[21]
Teeter, M. M., Yamano, A ., Stec, B. & Mohanty, U. On the nature of a glassy state of matter in a hydrated protein: Relation to protein function. Proceedings of the National Academy of Sciences 98, 11242–11247 (2001). 22. Li, J. et al. Reply to: Insucicient evidence for ageing in protein dynamics. Nat Phys 17, 775–776 (2021). 23. Ye, W. et al. Conformatio...
work page 2001
-
[37]
Subramaniam, S. & Kleywegt, G. J. A paradigm shift in structural biology. Nat Methods 19, 20–23 (2022). Figures: Figure 1: The proteins analyzed in this study. (A) Villin, the two segments for which the 𝛼-carbon RMSDs have been calculated are coloured differently. The 5rst segement comprising the 5rst 17 residue is green while the other is red. (B) CAP , ...
work page 2022
Reviewed August 10, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.