arxiv: 2604.16825 · v1 · submitted 2026-04-18 · ⚛️ physics.comp-ph

Recognition: unknown

Coarse-Grained Dynamics with Spatial Disorder and Non-Markovian Memory

Chuyi Liu , Yifeng Guan , Jingyuan Li , Mao Su

Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:20 UTC · model grok-4.3

classification ⚛️ physics.comp-ph

keywords coarse-grained dynamicsgeneralized Langevin equationspatial disorderanomalous diffusionnon-Markovian memoryvariational Bayesian inferenceheterogeneous systemsdata-driven modeling

0 comments

The pith

The spatial disorder-generalized Langevin equation separates fixed spatial variations from viscoelastic memory to extrapolate long-time anomalous diffusion from short trajectories in heterogeneous systems.

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

The paper develops a data-driven coarse-grained model for particle motion in systems where both position-dependent disorder and time-dependent friction are present. Conventional generalized Langevin equations that ignore spatial variations cannot recover the correct long-time behavior or ensemble averages when only short data are available. By contrast, the new formulation uses a Bayesian random-field approach to isolate the static disorder component, allowing the model to predict the crossover in diffusion scaling and the full statistical properties that arise from heterogeneity. If this separation holds, short simulation segments become sufficient to build reliable long-time descriptions without running full-scale computations. This matters for any system in which spatial inhomogeneity shapes observable transport, such as crowded media or materials with defects.

Core claim

The central claim is that standard generalized Langevin equations reach clear limits when spatial disorder is present, whereas the spatial disorder-generalized Langevin equation, constructed via variational Bayesian inference with a random field prior, disentangles static spatial disorder from non-Markovian friction and thereby accurately extrapolates long-time dynamics, captures the anomalous diffusion crossover, and recovers the ensemble statistical properties that are inherent to the disordered systems.

What carries the argument

The spatial disorder-generalized Langevin equation (SD-GLE), which employs a random field prior within variational Bayesian inference to represent position-dependent static disorder separately from the memory kernel that encodes viscoelastic friction.

If this is right

Standard generalized Langevin equations fail to capture long-time dynamics once spatial disorder is present.
SD-GLE extrapolates long-time trajectories and the anomalous diffusion crossover directly from short data segments.
The fitted model recovers the correct ensemble statistical properties that arise from the disordered nature of the system.
A data-driven coarse-grained description becomes feasible for heterogeneous systems without requiring a mean-field potential.

Where Pith is reading between the lines

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

The same separation technique could be tested on experimental single-particle tracking data where the underlying disorder is independently measurable.
If the random-field separation proves robust, many apparent memory effects observed in crowded media may turn out to be projections of static spatial heterogeneity rather than intrinsic time correlations.
The approach suggests a route to parameter-free upscaling: once disorder and memory are isolated from short runs, the resulting SD-GLE could serve as a surrogate for much longer simulations in porous or biological environments.

Load-bearing premise

The assumption that variational Bayesian inference with a random field prior can separate unchanging spatial disorder from time-dependent friction forces without creating artificial behaviors or losing physical consistency.

What would settle it

Generate an independent set of long trajectories in a known disordered system, fit the SD-GLE parameters exclusively from short segments of those trajectories, then simulate forward with the fitted model and check whether the predicted mean-squared-displacement crossover time and long-time ensemble variance match the direct long trajectories.

Figures

Figures reproduced from arXiv: 2604.16825 by Chuyi Liu, Jingyuan Li, Mao Su, Yifeng Guan.

**Figure 2.** Figure 2: FIG. 2. (a) Ensemble-averaged MSD generated from simu [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4. (a) Distribution of time-averaged MSDs for individ [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

read the original abstract

We introduce the spatial disorder-generalized Langevin equation (SD-GLE), a data-driven method for constructing coarse-grained (CG) dynamics in heterogeneous systems. Unlike conventional CG approaches that rely on a mean-field potential, SD-GLE utilizes a variational Bayesian framework with a random field prior to explicitly disentangle static spatial disorder from viscoelastic friction. Numerical results demonstrate the limits of standard GLEs, whereas SD-GLE accurately extrapolates long-time dynamics to capture the anomalous diffusion crossover from short trajectories and recover the ensemble statistical properties inherent to the disordered nature of these systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper adds a random-field prior inside variational Bayes to split static spatial disorder from the memory kernel in GLE coarse-graining, but the abstract-level claims rest on unshown numerics and leave identifiability untested.

read the letter

The central move is to replace the usual mean-field potential in GLE-based coarse-graining with a random field prior that is inferred variationally, so the model can attribute short-time fluctuations to either fixed spatial heterogeneity or viscoelastic memory and then extrapolate the crossover to anomalous diffusion at long times. That separation is the part that is not already in the standard GLE literature cited in the abstract, and it targets a real gap in heterogeneous soft-matter systems where ensemble statistics come from both sources at once. If the separation holds, it would let people build CG models from short trajectories that still recover the correct long-time ensemble behavior without having to simulate the full microscopic disorder everywhere. The abstract states that the numerics back this up and that ordinary GLEs fail where SD-GLE succeeds, which is the right kind of test to run. The approach is data-driven and stays within a Bayesian framework that can in principle quantify uncertainty, which is better than pure fitting. The soft spots are straightforward. No equations, no data description, no error bars, and no ground-truth test with known spatial statistics appear in the provided material, so the claim that the random field recovers the true disorder distribution cannot be checked. The stress-test concern about partial degeneracy is live: a slowly varying random field can produce correlation tails that look like memory, and nothing shown rules out that the variational posterior trades one against the other while still fitting the short data. Without an identifiability check or a controlled example where both contributions are known independently, the long-time extrapolation remains an assertion rather than a demonstrated result. This is for computational physicists and soft-matter people who already work with GLEs or variational methods for dynamics and need to handle explicit spatial disorder. A reader who wants a concrete extension of mean-field GLEs could extract the modeling idea, but only after the full derivations and validation controls are examined. It is worth sending to peer review so that experts can test the separation on known cases and ask for the missing metrics; the topic is relevant enough that the gaps can be fixed rather than rejected outright.

Referee Report

2 major / 3 minor

Summary. The manuscript introduces the spatial disorder-generalized Langevin equation (SD-GLE) as a data-driven coarse-graining approach for heterogeneous systems exhibiting both spatial disorder and non-Markovian memory. It employs a variational Bayesian framework with a random field prior to explicitly separate static spatial disorder from viscoelastic friction, in contrast to mean-field CG methods. Numerical results are presented to argue that standard GLEs fail to capture long-time behavior, while SD-GLE extrapolates anomalous diffusion crossovers from short trajectories and recovers the underlying ensemble statistics of the disordered system.

Significance. If the separation of disorder and memory proves reliable, the method offers a useful advance for coarse-graining in soft-matter and biological systems where spatial heterogeneity drives anomalous transport. The explicit use of a random-field prior within variational inference provides a principled way to quantify uncertainty and could enable predictions from limited short-time data. The data-driven character with Bayesian treatment of disorder is a clear strength relative to purely phenomenological GLE extensions.

major comments (2)

[Numerical Results] The central claim that SD-GLE 'accurately extrapolates long-time dynamics' and recovers ensemble statistics rests on the assumption that the variational posterior uniquely attributes fluctuations to either the random field or the memory kernel. No analysis of posterior identifiability or degeneracy (e.g., whether a slowly varying random field can trade off against a long-tailed memory kernel while fitting short trajectories) is provided; without ground-truth tests on synthetic data with known spatial statistics, the numerical demonstrations cannot rule out fitting artifacts.
[Methods] The variational Bayesian procedure is load-bearing for the entire method, yet the manuscript does not supply the explicit form of the random-field prior, the parameterization of the disorder field, or the evidence lower bound used in inference. Without these, it is impossible to verify that the inferred disorder field reproduces the true spatial correlation function when ground-truth heterogeneity is available.

minor comments (3)

[Abstract] The abstract asserts that 'numerical results demonstrate' the claimed accuracy but supplies no system details, trajectory lengths, error metrics, or comparison baselines; adding one quantitative sentence would strengthen the summary.
[Figures] Figure captions and axis labels in the numerical section should explicitly state the short-trajectory length used for inference versus the extrapolation horizon, and whether error bars represent posterior uncertainty or ensemble variability.
[Introduction] A brief comparison paragraph placing SD-GLE against existing Bayesian or random-field extensions of the GLE in the literature would clarify the incremental contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback and for recognizing the potential significance of the SD-GLE approach. We provide point-by-point responses to the major comments below and outline the revisions we will implement.

read point-by-point responses

Referee: [Numerical Results] The central claim that SD-GLE 'accurately extrapolates long-time dynamics' and recovers ensemble statistics rests on the assumption that the variational posterior uniquely attributes fluctuations to either the random field or the memory kernel. No analysis of posterior identifiability or degeneracy (e.g., whether a slowly varying random field can trade off against a long-tailed memory kernel while fitting short trajectories) is provided; without ground-truth tests on synthetic data with known spatial statistics, the numerical demonstrations cannot rule out fitting artifacts.

Authors: We appreciate the referee pointing out the need for explicit validation of identifiability. While our current numerical results use data from heterogeneous systems with known underlying disorder and memory properties to demonstrate extrapolation and recovery of ensemble statistics, we acknowledge that a dedicated identifiability study was not included. In the revised manuscript, we will add synthetic experiments specifically designed to test for degeneracy, such as comparing inference results when the random field correlation length is varied relative to the memory kernel timescale. This will include quantitative assessment of how well the inferred random field matches the true spatial statistics, thereby addressing concerns about fitting artifacts. revision: yes
Referee: [Methods] The variational Bayesian procedure is load-bearing for the entire method, yet the manuscript does not supply the explicit form of the random-field prior, the parameterization of the disorder field, or the evidence lower bound used in inference. Without these, it is impossible to verify that the inferred disorder field reproduces the true spatial correlation function when ground-truth heterogeneity is available.

Authors: We agree that the lack of these explicit details hinders full verification. The manuscript focuses on the conceptual framework and results, with some details in the supplementary information, but we will expand the main text or add a dedicated methods appendix in the revision. This will include the precise form of the random field prior (Gaussian random field with specified covariance function), the parameterization (e.g., via finite element or grid-based representation), and the derivation of the evidence lower bound (ELBO) used in the variational inference. With these additions, readers will be able to confirm the reproduction of spatial correlation functions on ground-truth data. revision: yes

Circularity Check

1 steps flagged

Long-time extrapolation presented as prediction but derived from short-trajectory variational fit

specific steps

fitted input called prediction [Abstract]
"SD-GLE accurately extrapolates long-time dynamics to capture the anomalous diffusion crossover from short trajectories and recover the ensemble statistical properties inherent to the disordered nature of these systems."

The variational Bayesian framework with random field prior is used to fit parameters to short trajectories; the long-time crossover and ensemble statistics are then generated from those same fitted parameters, so the 'accurate extrapolation' is produced by the model fit rather than tested against independent long-time data.

full rationale

The abstract describes a data-driven variational Bayesian method that fits an SD-GLE (with random-field prior) to short trajectories and then claims accurate long-time extrapolation of anomalous diffusion. This is a standard use of a fitted model for prediction rather than a mathematical derivation that reduces to its inputs by construction. No equations, self-citations, uniqueness theorems, or ansatzes are quoted that would trigger self-definitional or load-bearing circularity. The separation of disorder from memory is presented as the method's contribution, with numerical results offered as validation; without explicit held-out long-time tests or identifiability proofs in the visible text, the extrapolation claim carries some risk of being reconstruction, but does not meet the strict criteria for high circularity. The paper appears self-contained as a modeling proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

The central claim rests on the applicability of variational Bayesian inference and random field priors to separate disorder from memory; these are established tools but their effectiveness here is asserted without detailed parameter counts or independent validation in the abstract.

invented entities (1)

SD-GLE no independent evidence
purpose: Coarse-grained dynamics model that disentangles spatial disorder from non-Markovian friction
Introduced as the core new equation/framework in the abstract.

pith-pipeline@v0.9.0 · 5389 in / 1278 out tokens · 34590 ms · 2026-05-10T07:20:39.087923+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

49 extracted references

[1]

I. M. Toli ´ c Nørrelykke, E.-L. Munteanu, G. Thon, L. Oddershede, and K. Berg-Sørensen, Phys. Rev. Lett. 93, 078102 (2004)

2004
[2]

H. D. Pinholt, S. S.-R. Bohr, J. F. Iversen, W. Boomsma, and N. S. Hatzakis, Proceedings of the National Academy of Sciences118, e2104624118 (2021)

2021
[3]

I. M. Sokolov, Soft Matter8, 9043 (2012)

2012
[4]

P. Ge, Z. Zhang, and H. Lei, Phys. Rev. Lett.133, 077301 (2024)

2024
[5]

B. Mei, G. S. Grest, S. Liu, T. C. O’Connor, and K. S. Schweizer, Proceedings of the National Academy of Sci- ences121, e2403964121 (2024)

2024
[6]

B. Wang, S. M. Kuo, S. C. Bae, and S. Granick, Nature materials11, 481 (2012)

2012
[7]

G. Jung, G. Biroli, and L. Berthier, Phys. Rev. Lett.130, 238202 (2023)

2023
[8]

Bapst, T

V. Bapst, T. Keck, A. Grabska-Barwi´ nska, C. Donner, E. D. Cubuk, S. S. Schoenholz, A. Obika, A. W. R. Nel- son, T. Back, D. Hassabis, and P. Kohli, Nature Physics 16, 448 (2020)

2020
[9]

A. V. Weigel, B. Simon, M. M. Tamkun, and D. Krapf, Proceedings of the National Academy of Sciences of the United States of America108, 6438 (2011)

2011
[10]

Meroz, I

Y. Meroz, I. M. Sokolov, and J. Klafter, Phys. Rev. E 81, 010101 (2010)

2010
[11]

Goychuk and V

I. Goychuk and V. O. Kharchenko, Phys. Rev. Lett.113, 100601 (2014)

2014
[12]

Metzler, J.-H

R. Metzler, J.-H. Jeon, A. G. Cherstvy, and E. Barkai, Physical Chemistry Chemical Physics16, 24128 (2014)

2014
[13]

Meroz and I

Y. Meroz and I. M. Sokolov, Physics Reports573, 1 (2015)

2015
[14]

Thiel, F

F. Thiel, F. Flegel, and I. M. Sokolov, Phys. Rev. Lett. 111, 010601 (2013)

2013
[15]

Mu˜ noz-Gil, G

G. Mu˜ noz-Gil, G. Volpe, M. A. Garcia-March,et al., Na- ture Communications12, 6253 (2021)

2021
[16]

K. Song, D. E. Makarov, and E. Vouga, Phys. Rev. Res. 5, L012026 (2023)

2023
[17]

Mu˜ noz-Gil, G

G. Mu˜ noz-Gil, G. Volpe, M. A. Garcia-March, E. Aghion, A. Argun, C. Hong, T. Bland, S. Bo, J. A. Conejero, N. Firbas,et al., Nature communications12, 6253 (2021)

2021
[18]

Kowalek, H

P. Kowalek, H. Loch-Olszewska, and J. Szwabi´ nski, Phys- ical Review E100, 032410 (2019)

2019
[19]

S. Bo, F. Schmidt, R. Eichhorn, and G. Volpe, Physical Review E100, 010102 (2019)

2019
[20]

Tepper, B

L. Tepper, B. Dalton, and R. R. Netz, Journal of Chem- ical Theory and Computation20, 3061 (2024)

2024
[21]

H. Lei, N. A. Baker, and X. Li, Proceedings of the Na- tional Academy of Sciences113, 14183 (2016)

2016
[22]

J. O. Daldrop, B. G. Kowalik, and R. R. Netz, Phys. Rev. X7, 041065 (2017)

2017
[23]

Vroylandt, L

H. Vroylandt, L. Gouden` ege, P. Monmarch´ e, F. Pietrucci, and B. Rotenberg, Proceedings of the National Academy of Sciences119, e2117586119 (2022)

2022
[24]

Lyu and H

L. Lyu and H. Lei, Phys. Rev. Lett.131, 177301 (2023)

2023
[25]

P. Xie, R. Car, and W. E, Proceedings of the National Academy of Sciences121, e2308668121 (2024)

2024
[26]

Jiang and W

Y. Jiang and W. E, Journal of Chemical Theory and Computation20, 2824 (2024)

2024
[27]

Gouden` ege, T

L. Gouden` ege, T. Leli` evre, G. A. Pavliotis, and G. Stoltz, Multiscale Modeling & Simulation20, 1454 (2022)

2022
[28]

C. Ma, J. Wang, and W. E, Communications in Compu- tational Physics29, 1002 (2021)

2021
[29]

Lubelski, I

A. Lubelski, I. M. Sokolov, and J. Klafter, Phys. Rev. Lett.100, 250602 (2008)

2008
[30]

J.-H. Jeon, V. Tejedor, S. Burov, E. Barkai, C. Selhuber- Unkel, K. Berg-Sørensen, L. Oddershede, and R. Metzler, Phys. Rev. Lett.106, 048103 (2011)

2011
[31]

Akimoto, E

T. Akimoto, E. Barkai, and K. Saito, Phys. Rev. Lett. 117, 180602 (2016)

2016
[32]

Klimek, B

A. Klimek, B. A. Dalton, and R. R. Netz, The European Physical Journal E48, 55 (2025)

2025
[33]

Hij´ on, P

C. Hij´ on, P. Espa˜ nol, E. Vanden-Eijnden, and R. Delgado-Buscalioni, Faraday Discuss.144, 301 (2010)

2010
[34]

Bouchaud and A

J.-P. Bouchaud and A. Georges, Physics Reports195, 127 (1990)

1990
[35]

B¨ assler, Phys

H. B¨ assler, Phys. Rev. Lett.58, 767 (1987)

1987
[36]

Hecksher, A

T. Hecksher, A. I. Nielsen, N. B. Olsen, and J. C. Dyre, Nature Physics4, 737 (2008)

2008
[37]

M. Kong, L. Liu, X. Chen, K. I. Driscoll, P. Mao, S. B¨ ohm, N. M. Kad, S. C. Watkins, K. A. Bernstein, J. J. Wyrick, J.-H. Min, and B. Van Houten, Molecular Cell64, 376 (2016)

2016
[38]

Gerland, J

U. Gerland, J. D. Moroz, and T. Hwa, Proceedings of the National Academy of Sciences99, 12015 (2002)

2002
[39]

Slutsky, M

M. Slutsky, M. Kardar, and L. A. Mirny, Phys. Rev. E 69, 061903 (2004)

2004
[40]

B´ enichou, Y

O. B´ enichou, Y. Kafri, M. Sheinman, and R. Voituriez, Phys. Rev. Lett.103, 138102 (2009)

2009
[41]

Sheinman, O

M. Sheinman, O. B´ enichou, Y. Kafri, and R. Voituriez, Reports on Progress in Physics75, 026601 (2012)

2012
[42]

Titsias, inProceedings of the Twelfth Interna- tional Conference on Artificial Intelligence and Statis- tics, Vol

M. Titsias, inProceedings of the Twelfth Interna- tional Conference on Artificial Intelligence and Statis- tics, Vol. 5, edited by D. van Dyk and M. Welling (PMLR, 2009) pp. 567–574

2009
[43]

Ceriotti, G

M. Ceriotti, G. Bussi, and M. Parrinello, Journal of Chemical Theory and Computation6, 1170 (2010)

2010
[44]

Ceriotti, G

M. Ceriotti, G. Bussi, and M. Parrinello, Phys. Rev. Lett. 102, 020601 (2009)

2009
[45]

Schweppe, IEEE Transactions on Information Theory 11, 61 (1965)

F. Schweppe, IEEE Transactions on Information Theory 11, 61 (1965)

1965
[46]

Cammarota and G

C. Cammarota and G. Biroli, Proceedings of the National Academy of Sciences109, 8850 (2012)

2012
[47]

Chakrabarty, R

S. Chakrabarty, R. Das, S. Karmakar, and C. Dasgupta, The Journal of Chemical Physics145, 034507 (2016)

2016
[48]

Rahman, Phys

A. Rahman, Phys. Rev.136, A405 (1964)

1964
[49]

Y. He, S. Burov, R. Metzler, and E. Barkai, Phys. Rev. Lett.101, 058101 (2008)

2008