pith. sign in

arxiv: 2604.18068 · v1 · submitted 2026-04-20 · ❄️ cond-mat.stat-mech · cond-mat.soft

Self-averaging parameter estimation for coarse-grained particle models

Carlos Monago , J. A. de la Torre , Pep Espa\~nol This is my paper

Pith reviewed 2026-05-10 03:45 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech cond-mat.soft
keywords parameter estimationcoarse-grained modelsstochastic differential equationsself-averaginghydrodynamic mobilityLennard-Jones fluidpotential of mean forcemobility tensor
0
0 comments X

The pith

Coupling a coarse-grained SDE to its own parameter equations produces self-averaging that matches selected microscopic and mesoscopic observables.

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

The paper introduces a parameter estimation technique that takes averages and correlations of chosen observables from microscopic data and uses them to determine both equilibrium and dynamic parameters in a coarse-grained stochastic differential equation. The method augments the SDE with evolution equations for the parameters themselves, creating a joint dynamical system whose long-time behavior drives the parameters to the values that reproduce the microscopic statistics at the mesoscopic scale. Validation on a Brownian particle in a harmonic potential and on hydrodynamically interacting particles with the Rotne-Prager-Yamakawa tensor shows close agreement, after which the technique is applied to extract the potential of mean force and position-dependent mobility tensor for heavy particles in a bimodal-mass Lennard-Jones fluid. A sympathetic reader would care because the procedure supplies a systematic, non-manual route to consistent coarse-grained models that can handle state-dependent transport coefficients.

Core claim

The paper claims that the coupled system consisting of the coarse-grained stochastic differential equation with free parameters and the auxiliary dynamic equations for those parameters self-averages, according to Anosov-Kifer's theorem, so that the final parameter values make the averages and correlations of selected observables coincide between the microscopic and mesoscopic descriptions.

What carries the argument

The central mechanism is the self-averaging property of the augmented dynamical system formed by the coarse-grained SDE together with the parameter-evolution equations, which enforces observable matching without external optimization.

If this is right

  • The method determines both static parameters appearing in the reversible part of the dynamics, such as those in the free-energy function or potential of mean force, and dynamic parameters such as friction coefficients.
  • It identifies state-dependent transport properties, for instance the position-dependent form of the mobility tensor for hydrodynamically interacting particles.
  • Application to a bimodal-mass Lennard-Jones fluid infers both the potential of mean force between the heavy particles and its hydrodynamic mobility tensor.
  • The procedure is not restricted to constant parameters and extends directly to position-dependent coefficients.

Where Pith is reading between the lines

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

  • If the self-averaging convergence holds for a broader class of observables, the technique could be used to derive coarse-grained models for molecular systems without presupposing functional forms for interactions or transport.
  • One could test convergence rates by applying the method to analytically solvable coarse-graining problems and measuring how quickly parameters stabilize.
  • The dynamical adjustment of parameters might connect to other adaptive model-reduction approaches by offering an explicit time-evolution route to consistency between scales.

Load-bearing premise

The coupled system of the coarse-grained stochastic differential equation and the parameter-evolution equations self-averages to parameter values that make the chosen microscopic observables coincide with their mesoscopic counterparts.

What would settle it

Running the coupled system for the Brownian particle in a harmonic potential and verifying whether the converged friction and spring-constant parameters reproduce the known microscopic values within numerical error would falsify the claim if they fail to match.

Figures

Figures reproduced from arXiv: 2604.18068 by Carlos Monago, J. A. de la Torre, Pep Espa\~nol.

Figure 1
Figure 1. Figure 1: FIG. 1: Langevin particle in a harmonic trap in three dynamical regimes. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Validation of the parameterized mobility model. Vertical dashed line corresponds to [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Validation of the parameterized mobility model. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Fitted building blocks at two solvent number densities [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Radial distribution function [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 5
Figure 5. Figure 5: Both models—with and without hydrodynamic [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Comparison between the HI model and the Stokes model against MD results for the distance-resolved [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Velocity autocorrelation function (VACF) [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Comparison of the HI and Stokes models with [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

We introduce a parameter estimation method that utilizes microscopic data, specifically averages and correlations of selected microscopic observables, to determine the parameters of a stochastic differential equation governing coarse-grained degrees of freedom. The method is not limited to static parameters found in the reversible part of the coarse-grained dynamics, such as those in the free energy function or potential of mean force, but also extends to dynamic parameters, including friction coefficients. The method couples the stochastic differential equation with free parameters to dynamic equations for the parameters. The coupled system self-averages, according to Anosov-Kifer's theorem, in such a way that the final state of the parameters gives coincidence between the microscopic and mesoscopic averages and correlations of selected observables. The method is validated in two examples: a Brownian particle in a harmonic potential, and a set of Brownian particles interacting hydrodynamically with the Rotne-Prager-Yamakawa mobility tensor. This latter case illustrates how the method can be used not only to determine coefficients but also state dependent transport properties - in this case, the position dependent form of the mobility tensor. The parameter estimation for these two models yields excellent results. Subsequently we use the methodology to study a bimodal-mass Lennard-Jones fluid for which we infer both the potential of mean force between the heavy particles and its hydrodynamic mobility tensor.

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.

Referee Report

2 major / 2 minor

Summary. The paper introduces a parameter estimation method for coarse-grained stochastic differential equations (SDEs) by coupling the SDE (with free parameters) to auxiliary dynamic equations for those parameters. The coupled system is claimed to self-average, per Anosov-Kifer's theorem, such that the parameters converge to values making selected microscopic averages and correlations coincide with their mesoscopic counterparts. The approach is validated on a Brownian particle in a harmonic potential and on hydrodynamically interacting particles using the Rotne-Prager-Yamakawa (RPY) mobility tensor (including inference of its position-dependent form); it is then applied to a bimodal-mass Lennard-Jones fluid to recover both the potential of mean force between heavy particles and the associated hydrodynamic mobility tensor.

Significance. If the self-averaging convergence holds, the method supplies a systematic, non-optimization-based route to infer both static (free-energy/PMF) and dynamic (friction/mobility) parameters from microscopic data. The extension to state-dependent transport coefficients is a clear strength, and the use of a rigorous theorem for justification distinguishes it from purely heuristic matching schemes.

major comments (2)
  1. [Abstract and validation sections] Abstract and validation sections: the claim that 'the parameter estimation for these two models yields excellent results' is unsupported by any quantitative error metrics, baseline comparisons, or tabulated discrepancies between microscopic and mesoscopic observables; without these, the validation of the central claim cannot be assessed.
  2. [Theoretical framework] Theoretical framework (invocation of Anosov-Kifer theorem): the manuscript applies the theorem to the joint (coarse-grained SDE + parameter-evolution) system for the RPY and bimodal-mass LJ cases but supplies no verification that the required strong mixing, hyperbolicity, or uniform timescale-separation conditions hold for the chosen observables and parameter regimes; if these hypotheses fail, the parameter flow need not converge to the desired matching point.
minor comments (2)
  1. [Abstract and results sections] The abstract and results sections do not specify which microscopic observables (averages/correlations) were selected for matching, nor do they report the final converged parameter values or their uncertainties.
  2. No discussion is provided of how the method behaves when the timescale separation is only marginal or when the microscopic dynamics exhibit weak mixing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive evaluation of the method's significance and for the detailed major comments, which help clarify the presentation of our results. We address each point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract and validation sections] Abstract and validation sections: the claim that 'the parameter estimation for these two models yields excellent results' is unsupported by any quantitative error metrics, baseline comparisons, or tabulated discrepancies between microscopic and mesoscopic observables; without these, the validation of the central claim cannot be assessed.

    Authors: We agree that quantitative support is needed to substantiate the validation claims. In the revised manuscript, we will expand the validation sections for both the harmonic oscillator and RPY models by adding tabulated metrics, including relative errors and L2 discrepancies for key observables (means, variances, and time correlations) between the microscopic targets and the converged mesoscopic values. We will also include a baseline comparison against a direct least-squares optimization of the same observables to quantify the performance of the self-averaging approach. These additions will enable readers to rigorously assess the accuracy of the parameter estimates. revision: yes

  2. Referee: [Theoretical framework] Theoretical framework (invocation of Anosov-Kifer theorem): the manuscript applies the theorem to the joint (coarse-grained SDE + parameter-evolution) system for the RPY and bimodal-mass LJ cases but supplies no verification that the required strong mixing, hyperbolicity, or uniform timescale-separation conditions hold for the chosen observables and parameter regimes; if these hypotheses fail, the parameter flow need not converge to the desired matching point.

    Authors: The referee correctly identifies that the manuscript invokes Anosov-Kifer averaging without explicit verification of the theorem's hypotheses. While the underlying Brownian and Langevin dynamics are expected to satisfy strong mixing and hyperbolicity (as established in the stochastic dynamics literature for such systems), we did not provide case-specific checks or timescale-separation diagnostics. In revision, we will add a subsection to the theoretical framework discussing the applicability of the conditions, supported by references on ergodicity of Langevin equations and numerical evidence from our simulations (e.g., autocorrelation decay times demonstrating mixing and separation). A full rigorous proof of hyperbolicity for the high-dimensional LJ case lies beyond the scope of this applied work, but the added discussion will strengthen the justification. revision: partial

Circularity Check

0 steps flagged

No significant circularity; convergence justified by external theorem rather than self-definition

full rationale

The paper's central construction couples a coarse-grained SDE (with free parameters) to auxiliary evolution equations for those parameters. It then invokes Anosov-Kifer's theorem—an external result in dynamical systems—to conclude that the joint system self-averages so the parameters converge to values making chosen microscopic and mesoscopic observables coincide. This coincidence is the designed target of the auxiliary equations, yet the justification for why the dynamics reach that target is supplied by the cited theorem rather than by redefining the parameters in terms of the match itself or by a self-citation chain. Validation examples recover known parameters and the LJ application infers new ones; neither step reduces the claimed result to its inputs by construction. The derivation therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the applicability of Anosov-Kifer's theorem to the coupled system; no additional free parameters or invented entities are introduced by the method itself.

axioms (1)
  • standard math Anosov-Kifer's theorem guarantees self-averaging of the coupled system so that parameters converge to values reproducing the selected microscopic averages and correlations.
    Invoked to justify convergence of the parameter dynamics to the matching state.

pith-pipeline@v0.9.0 · 5540 in / 1418 out tokens · 43527 ms · 2026-05-10T03:45:56.798190+00:00 · methodology

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

40 extracted references · 40 canonical work pages

  1. [1]

    These are the building blocks of the CG model (51) describing the dynamics of the heavy particles

    The parameterized CG model In Fig 4 we show the pair-wise potential of mean force (panel (a)) and the pair mobilities (panel (b)) that are obtained after convergence of the evolution of the pa- rameters according to the coupled evolution of (51) and (59). These are the building blocks of the CG model (51) describing the dynamics of the heavy particles. Ob...

    work page

  2. [2]

    4, we compare several macroscopic observables computed in both the microscopic MD sim- ulations and the CG model

    Validation of the parameterized CG model To validate the model (51) with the fitted building blocks shown in Fig. 4, we compare several macroscopic observables computed in both the microscopic MD sim- ulations and the CG model. The first observable we consider is the RDF shown in Fig. 5. Both models—with and without hydrodynamic interactions—produce indis...

    work page

  3. [3]

    Range of lag times The coupled dynamics (51),(59) leads to the equality of correlations, as in (25), for a particular value of the lag time∆t=τ S, τ⊥, τ∥. If the parameterized Markovian model is a good description of the underlying microscopic behaviour, weshouldobtainasetofparametersµ 0, aα, bα that do not depend on the value of the lag time∆tused to fit...

    work page 2000

  4. [4]

    Example 1: Brownian trap For the harmonic Langevin oscillator discussed in Sec. V, increasing the damping coefficientγreduces the momentum correlation at any fixed lagτ >0, so that ∂ ∂γ Cγ(τ)<0.(A9) Applying the stability rule (A8) therefore yields dγ(t) dt =− 1 Tparam [Cp(τ)−p(t−τ)p(t)].(A10) If the simulated correlation is too large compared with the ta...

    work page

  5. [5]

    Example 2: Brownian trap with HI In the example of Sec. VI we consider the observables X∈ {O, G α, Hα}with mesoscopic predictions CX(τ;p)≡ ⟨X(t, τ)⟩ p .(A11) The parameters ˜D0,a α, andb α enter the mobility tensor D(R)and therefore control the magnitude of the dis- placement increments∆R µ(t, τ). Increasing any of these mobility coefficients enhances the...

    work page

  6. [6]

    Example 3: LJ binary mixture In the LJ mixture considered in Sec. VII, the irre- versible dynamics is controlled by the mobility matrix µ(R), while the friction tensor is its inverse,Γ(R) = µ(R)−1.The self parameterµ 0 is fitted through the lagged displacement observableO(t, τ S), whereas the spline coefficientsa α andb α parameterize the transverse and l...

    work page

  7. [7]

    Unless otherwise stated, the simulation parameters areγ ∗ =k BT=m= 1and dt= 0.001

    Example 1: Brownian trap The stochastic dynamics (26) is integrated using the Grønbech-Jensen and Farago (G-JF) algorithm [30], rec- ommended in [31], while the parameter evolution equa- tions (30) are integrated with a first-order Euler scheme due to the large value ofTparam. Unless otherwise stated, the simulation parameters areγ ∗ =k BT=m= 1and dt= 0.0...

    work page

  8. [8]

    Example 2: Brownian trap with HI For the ground truth (GT) simulations we chose a sys- tem of 100 Brownian particles under the action of the RPY tensor (36). This number of particles was chosen in order to obtain correct statistics, while keeping in mind that the Cholesky decomposition of the RPY tensor be- comes increasingly expensive as the number of pa...

    work page

  9. [9]

    We considered four tracer massesm B ∈ {2,5,10,50}

    Example 3: LJ binary mixture All LAMMPS runs were carried out at temperature T= 2and number densitiesρ= 0.6andρ= 0.8 in a periodic cubic box of side lengthL= 25.6285, containingN A = 10 000light particles (m A = 1) and NB = 100heavy tracers. We considered four tracer massesm B ∈ {2,5,10,50}. All particles, light and heavy, interact with the same LJ potent...

    work page

  10. [10]

    Einstein, Ann

    A. Einstein, Ann. Phys. (Leipzig)19, 549 (1905)

    work page 1905

  11. [11]

    Green, J

    M. Green, J. Chem. Phys.20, 1281 (1952)

    work page 1952

  12. [12]

    Zwanzig, Physical Review124, 983 (1961)

    R. Zwanzig, Physical Review124, 983 (1961)

    work page 1961

  13. [13]

    W. G. Noid, The Journal of Physical Chemistry B127, 4174 (2023)

    work page 2023

  14. [14]

    A. P. Lyubartsev and A. Laaksonen, Physical Review E 52, 3730 (1995)

    work page 1995

  15. [15]

    Izvekov and G

    S. Izvekov and G. A. Voth, The Journal of Physical Chemistry B109, 2469 (2005)

    work page 2005

  16. [16]

    W. G. Noid, J.-W. Chu, G. S. Ayton, V. Krishna, S. Izvekov, G. a Voth, A. Das, and H. C. Andersen, J. Chem. Phys.128, 244114 (2008)

    work page 2008

  17. [17]

    M.S.Shell,TheJournalofChemicalPhysics129,144108 (2008)

    work page 2008

  18. [18]

    Papavasiliou, G

    A. Papavasiliou, G. Pavliotis, and A. Stuart, Stochastic Processes and their Applications119, 3173 (2009)

    work page 2009

  19. [19]

    Hijón, P

    C. Hijón, P. Español, E. Vanden-Eijnden, and R.Delgado-Buscalioni,FaradayDiscuss.144,301(2010)

    work page 2010

  20. [20]

    Dequidt and J

    A. Dequidt and J. G. Solano Canchaya, The Journal of Chemical Physics143, 084122 (2015)

    work page 2015

  21. [21]

    Y. Han, J. Jin, and G. A. Voth, The Journal of Chemical Physics154, 084122 (2021)

    work page 2021

  22. [22]

    V. P. Sokhan and I. T. Todorov, Molecular Simulation 47, 248 (2021)

    work page 2021

  23. [23]

    Abdulle, G

    A. Abdulle, G. Garegnani, G. A. Pavliotis, A. M. Stuart, and A. Zanoni, Foundations of Computational Mathe- matics23, 33 (2023)

    work page 2023

  24. [24]

    Milster, J

    S. Milster, J. Dzubiella, G. Stock, and S. Wolf, The Jour- nal of Chemical Physics162, 154113 (2025)

    work page 2025

  25. [25]

    Dietrich, A

    F. Dietrich, A. Makeev, G. Kevrekidis, N. Evangelou, T. Bertalan, S. Reich, and I. G. Kevrekidis, Chaos: An Interdisciplinary Journal of Nonlinear Science33, 023121 (2023)

    work page 2023

  26. [26]

    T. Ye, B. Jing, and D. Pan, Journal of Computational Physics475, 111857 (2023)

    work page 2023

  27. [27]

    Sachs, W

    M. Sachs, W. G. Stark, R. J. Maurer, and C. Ortner, Machine Learning: Science and Technology6, 015016 (2025)

    work page 2025

  28. [28]

    Kifer, Stochastics and Dynamics01, 1 (2001)

    Y. Kifer, Stochastics and Dynamics01, 1 (2001)

    work page 2001

  29. [29]

    Monago, J

    C. Monago, J. A. D. L. Torre, R. Delgado-Buscalioni, and P. Español, The Journal of Chemical Physics162, 114115 (2025)

    work page 2025

  30. [30]

    T. G. Mason, K. Ganesan, J. H. van Zanten, D. Wirtz, and S. C. Kuo, Phys. Rev. Lett.79, 3282 (1997)

    work page 1997

  31. [31]

    T. A. Waigh, Reports on Progress in Physics68, 685 (2005)

    work page 2005

  32. [32]

    Bocquet, J.-P

    L. Bocquet, J.-P. Hansen, and J. Piasecki, Journal of Sta- tistical Physics89, 321 (1997)

    work page 1997

  33. [33]

    Reith, H

    D. Reith, H. Meyer, and F. Müller-Plathe, Macro- molecules34, 2335 (2001), arXiv:cond-mat/0008338

    work page arXiv 2001

  34. [34]

    Press, S

    W. Press, S. Teukolski, W. Vetterling, and B. Flannery, Numerical Recipes in Fortran 77(Cambridge University Press, 1992)

    work page 1992

  35. [35]

    Givon, R

    D. Givon, R. Kupferman, and A. Stuart, Nonlinearity 17, R55 (2004)

    work page 2004

  36. [36]

    D. A. Beard and T. Schlick, The Journal of Chemical Physics112, 7313 (2000)

    work page 2000

  37. [37]

    Tanygin and S

    B. Tanygin and S. Melchionna, Computer Physics Com- munications299, 109152 (2024)

    work page 2024

  38. [38]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolin- tineanu, W. M. Brown, P. S. Crozier, P. J. In ’T Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton, Computer Physics Communications271, 108171 (2022)

    work page 2022

  39. [39]

    Grønbech-Jensen, N

    N. Grønbech-Jensen, N. R. Hayre, and O. Farago, Com- puter Physics Communications185, 524 (2014)

    work page 2014

  40. [40]

    Finkelstein, G

    J. Finkelstein, G. Fiorin, and B. Seibold, Molecular Physics118, e1649493 (2020)

    work page 2020