Pith. sign in

REVIEW 5 major objections 5 minor 3 cited by

Many-Body Coarse-Grained Molecular Dynamics with the Atomic Cluster Expansion

T0 review · 5 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read The paper claims that the Atomic Cluster Expansion can model the many-body potential of mean force in coarse-grained molecular dynamics, with body-order-controlled accuracy and linear cost.

desk verdict Applies ACE to CG force matching with a clean body-order study; the results look credible, but the main loss equation is garbled and needs correction before the method is reproducible. read the letter →

arxiv 2502.04661 v1 pith:MWSYAK65 submitted 2025-02-07 physics.comp-ph

classification physics.comp-ph
keywords coarse-grainedmoleculardynamicsatomicclusterexpansionpotentialofmeanforcemany-bodyinteractionsmatchingradialdistributionfunctionangularstarpolymers
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to show that the Atomic Cluster Expansion, a systematic many-body basis for interatomic potentials, also works to represent the potential of mean force in coarse-grained molecular dynamics. On star-polymer fluids and methanol, force-matched ACE models reproduce the all-atom radial and angular distribution functions, and accuracy improves as the body order is raised from two to four. The strongest demonstration is the polymer case: body order 4 cuts the RDF error from 4.711 (pairwise) to 0.058, and ADF errors fall by an order of magnitude. If the claim holds, accurate CGMD no longer needs bespoke pair potentials or expensive constrained simulations, because a single linear basis fitted to ordinary MD trajectories is systematically improvable and scales linearly with system size.

What carries the argument

The central object is the Atomic Cluster Expansion parameterization of the coarse-grained potential: a linear basis of site energies $U_I^{\mathrm{ACE}}(\theta; R,\zeta)$ that is invariant under relabeling, rotations, and reflections, built from tensor products of one-particle basis functions (radial polynomials times spherical harmonics) and symmetrized with Clebsch–Gordan couplings, then truncated by correlation order $\nu_{\max}$ and total degree $D_{\mathrm{tot}}$. This gives a systematically improvable, size-extensive model whose cost is linear in particle number and neighbor count. The parameters are fixed by force matching: a least-squares loss comparing ACE forces to instantaneous collective forces sampled from unconstrained atomistic MD, justified by the equivalence of this loss with exact mean-force matching.

What would settle it

Measure the residual mean force between two coarse-grained sites in the star-polymer melt separated just beyond the 16 Å cutoff, using constrained atomistic simulations; a systematic non-zero force would show the locality assumption fails, and ACE-CG's success at that cutoff would then be specific to this system rather than general.

Watch

Extended reading notes

Core claim

The paper's central claim is that the potential of mean force governing coarse-grained sites—a conditional free energy obtained by integrating out atomic fluctuations—is well approximated by an Atomic Cluster Expansion potential with a finite cutoff and a modest body order. In the two test systems the claim is quantitative: for star polymers, raising the body order from two to four lowers the RDF error from $E_{\mathrm{RDF}}=4.711$ to $0.058$ and reduces the ADF error by a factor of 4–13 depending on the angular cutoff; for methanol, the two-site body-order-4 model reaches $E_{\mathrm{RDF}}=0.005$ and, at $r_{\mathrm{cut}}=8$ Å, $E_{\mathrm{ADF}}=0.010$, down from $0.072$ at body order two. The authors interpret these numbers as evidence that many-body terms, not pairwise refinements, carry the missing angular and packing information in these fluids.

Load-bearing premise

The load-bearing premise is that the true coarse-grained forces are local: a site's interaction with other sites dies out within the chosen cutoff, with no long-range effect from the averaged-out internal motions.

Editorial extensions

If this is right

  • Coarse-grained simulations with ACE-CG can target quantitative equilibrium structure, not just qualitative behavior, at large speedups (e.g., 61–108 ns/day vs 1.3 ns/day for the polymer melt).
  • Systematic body-order convergence gives a practical dial: if a pairwise CG model fails, the remedy is to raise the body order rather than redesign the potential representation.
  • Force matching on instantaneous forces makes training cheap, because standard unconstrained MD trajectories suffice and the noisy collective forces share the same minimizer as exact mean forces.
  • Models trained on small cells transfer to larger systems at the same density with a small accuracy loss (methanol two-site RDF error 0.005 vs 0.003 size-consistent).
  • Pairwise-only representations, including iterative Boltzmann inversion, can reproduce RDFs but miss angular structure; many-body terms are necessary for accurate ADFs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Going beyond the paper, the locality caveat implies that ACE-CG should be tested on systems with long-ranged electrostatics or strong density and temperature gradients, where the neglected nonlocal part of the potential of mean force is more likely to matter.
  • Because the ACE basis is complete as body order, cutoff, and degree grow, the same fitting pipeline could plausibly be extended to anisotropic coarse-grained sites or to memory kernels, directions the paper names as future work.
  • The methanol ADF results suggest that most of the missing angular correlation is carried by three-body terms (the jump from body order 3 to 4); ablating specifically that contribution would be a natural diagnostic.
  • The potential of mean force is temperature dependent, so a model fitted at one temperature may not transfer; the paper does not test this, making temperature-transferability a useful next experiment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper introduces ACE-CG, a coarse-grained molecular dynamics approach in which the potential of mean force is represented by the Atomic Cluster Expansion as a truncated many-body site expansion, with parameters fitted by instantaneous-force matching. The method is tested on star-polymer fluids and methanol (one-site and two-site mappings), comparing body orders 2, 3, and 4. The authors report that body order 4 gives the best agreement with all-atom reference RDFs and ADFs, outperforming lower body orders and Iterative Boltzmann Inversion in most comparisons, and they report large computational speedups. A transferability test from 50 to 400 methanol molecules is also presented.

Significance. If the reported results hold, ACE-CG offers a systematic, interpretable, and computationally efficient route to many-body coarse-grained potentials, with a clear demonstration that increasing body order improves structural accuracy. The paper's strengths include the use of a standard force-matching principle, a genuine forward validation of RDFs and ADFs that are not directly fitted, a size-transferability test for methanol, and direct comparisons against IBI and lower body orders. The numerical evidence for the central claim is substantial. However, the manuscript as written contains an inconsistency between the main-text definition of the instantaneous collective force and the derivation in Appendix A.2, and Eq. (2.10) as written is not a well-defined loss. This must be resolved before the reported results can be reproduced or fully trusted.

major comments (5)
  1. [Section 2.3, Eq. (2.10), Appendix A.2] The definition of the instantaneous collective force (ICF) is internally inconsistent. In Section 2.3 the ICF is defined as the weighted sum F_I(r,z)=sum_i w_Ii f_i(r,z) using the COM weights of Eq. (2.1), but Appendix A.2 defines it as the unweighted sum over atoms in the CG site, F_I(r,z)=sum_{j in J_I} f_j, and derives Eq. (A.9). These two definitions are not equivalent for a general linear mapping with non-uniform weights; the conditional average of the weighted sum is not the mean force conjugate to R_I, so the two choices lead to different fitted potentials. In addition, Eq. (2.10) compares the atomic gradient grad_r U^ACE (a 3n-dimensional vector) with F(r,z) (an N-dimensional vector of site forces), so the norm in the loss is not defined. The intended loss is presumably ||grad_R U^ACE + F_ICF||^2, but as written the fitting procedure cannot be identified from the manuscript. Since the central numerical results depend on this loss, please specify the exact loss and ICF convention, and if the implementation follows Appendix A.2, correct Section 2.3 and Eq. (2.10) accordingly.
  2. [Tables 1, 2, 4, 5] The reference MD curves include error bars from five repeated simulations, but the ACE-CG and IBI results are reported without any uncertainty quantification. The central claim of systematically improvable accuracy rests on differences such as ERDF = 0.152 (BO3) versus 0.058 (BO4) for the polymer system, and on the methanol ADF drop from 0.072 (BO2) to 0.010 (BO4). Without error bars or multiple-seed estimates for the CG simulations, it is unclear whether these differences are significant or whether they could vary with the random seed, the number of training frames, or the simulation length. Please provide statistical uncertainties for the reported CG error metrics, at least for the headline comparisons.
  3. [Appendix B, Section 3.1] The hyperparameters listed in Table 7 (nu_max, D_max, rcut) are not sufficient to reproduce the fits. The manuscript does not state how many atomistic configurations were used for training, how many force samples per configuration, how the radial basis and regularization were set beyond the smoothness prior p=4, or how the data sets for the size-consistent and size-transferable methanol models were constructed. Since the force-matching loss is the core of the method, a complete and reproducible specification of the training data and fitting procedure is necessary. Please provide this information, ideally with a repository of the fitting scripts and training data.
  4. [Section 3.2.2, Table 1] The IBI comparison is stopped after six iterations with alpha=0.2, and no convergence criterion is given. The statement that ACE-CG BO4 is more accurate than IBI on the polymer RDF (ERDF 0.058 vs 0.183) may therefore reflect an unconverged IBI run rather than a fundamental limitation of pair potentials for the RDF. The ADF comparison is a fairer comparison because IBI is inherently pairwise, but the RDF claim should either be qualified or IBI should be run to convergence. Please also report the IBI RDF error after more iterations or state why six iterations are sufficient.
  5. [Section 2.2, Eq. (2.6)] The locality assumption behind the finite cutoff Rcut is acknowledged in the text as lacking a theoretical justification. This is an honest and appropriate statement, and the empirical support is reasonable for the two tested systems. However, the paper does not study the sensitivity of the results to Rcut (e.g., by varying 16 Å for polymers or 6-7 Å for methanol), so the robustness of the conclusions with respect to this load-bearing assumption is not demonstrated. A brief cutoff-convergence study, even for one system, would materially strengthen the central claim.
minor comments (5)
  1. [Eq. (3.6), Table 4] The error metric in Eq. (3.6) is defined for a single radial distribution function, but the two-site methanol results in Table 4 involve multiple site-site RDFs. Please specify how the reported two-site ERDF value is computed (e.g., averaged over C-C and O-O, or summed).
  2. [Section 3.2.1] The symbol N is used for both the total number of atoms in Eq. (2.5) and the number of CG particles in Eq. (2.6), while Section 3.2.1 says 'N = 265 polymer molecules'. This overloading of notation can confuse the reader; please use distinct symbols for the number of molecules, atoms, and CG sites.
  3. [Section 2.3, first paragraph] The sentence 'The ICF is identical to the local mean force employed in the adaptive biasing force method' is not correct for the weighted ICF defined just above it. The ABF local mean force for a COM mapping is the total force on the group, not the mass-weighted sum. This statement should be corrected together with the ICF definition.
  4. [Appendix A.2, Eq. (A.8)] The notation in Eq. (A.8) omits the dependence of F_I on R and uses the notation F_I(ζ); this is likely a typographical issue but worth correcting for clarity.
  5. [Table 3, Table 6] The bracketed 'ns/day' values in Tables 3 and 6 are based on optimized computational kernels that are 'not yet available through ASE or LAMMPS'. Please state explicitly whether these values are projected estimates or measured with a different interface, and clarify the basis for the speedup factors.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the RDF/ADF results are genuine forward predictions from a force-matched model, and the ACE-related self-citations are not load-bearing in a circular sense.

full rationale

The central numerical claims are not circular. The ACE-CG parameters are fitted by force matching to instantaneous atomic forces (Eq. 2.10 and Appendix A.4), not to the equilibrium RDF or ADF that are later compared with all-atom MD. The RDFs and ADFs are obtained by running independent CG simulations with the fitted potential, so they are genuinely predicted observables; this is reinforced by the transfer from a 50-molecule training box to a 400-molecule test box and by the systematic decrease of ERDF and EADF with increasing body order. The ACE completeness statement cited from the authors' own earlier work [2] is a mathematical approximation property of the basis, and it is used to justify expressivity, not to fit any parameter; the paper explicitly states that many alternative architectures could be used for the same task (Section 2.2), so no uniqueness claim is being imported to forbid alternatives. The use of ACEpotentials.jl [71] is an implementation choice, not a fitted input. The locality assumption is stated honestly as lacking theoretical justification and supported only by empirical studies, so it is not smuggled in. The one notable issue is an internal inconsistency in Section 2.3: the main text defines the instantaneous collective force as the weighted sum FI = sum_i w_Ii f_i, while Appendix A.2 defines it as the unweighted sum over atoms in the site and derives Eq. (A.9); additionally Eq. (2.10) compares grad_r U^ACE with an N-dimensional site force F(r,z), which is dimensionally ill-posed as written. This is a correctness and reproducibility flaw in the written method, not a circularity, because the reported RDF/ADF predictions are not equivalent to the force-matching target by construction. The score of 1 reflects only the presence of minor self-citations for the ACE basis and implementation; these citations do not make the derivation circular.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim relies on the standard potential-of-mean-force framework, on ACE completeness results cited from the authors' own earlier work, and on the equivalence between mean-force and instantaneous-force losses attributed to DeepCG. The only novel fitted objects are the ACE coefficients theta, which are regularized linear regression weights; they are numerous and not enumerated individually. The hyperparameters (Rcut, Dmax, nu_max, smoothness prior p) are chosen by hand and reported in Table 7 and Appendix B, so they are listed as free parameters. No new physical entities are introduced.

free parameters (5)
  • cutoff radius Rcut = 16.0 (polymer); 7.0 (methanol one-site); 6.0 (methanol two-site)
    Chosen by hand (Table 7); controls which neighbors contribute to the site energy. The paper states there is no theoretical justification for locality, so this choice is part of the model.
  • total degree Dmax = 22/18/16 for polymer BO2/3/4; 18/14/14 for methanol one-site; 14/12/12 for two-site
    Selected per model (Table 7) to balance expressivity and cost; not optimized by data.
  • correlation order nu_max (body order) = 1, 2, 3 (body orders 2, 3, 4)
    The model family parameter varied to test many-body effects; chosen deliberately rather than fitted.
  • Tikhonov smoothness prior p = 4
    Chosen in Appendix B (p=4) to regularize the linear fit; affects smoothness of the resulting potential.
  • IBI update factor alpha = 0.2
    Set in Appendix A.3; only affects the IBI baseline comparison, not the central ACE-CG result.
assumptions (5)
  • domain assumption Loss equivalence: the minimizers of the mean-force loss (2.8) and the instantaneous-force loss (2.10) coincide
    Invoked in Section 2.3 and attributed to reference 75; the paper does not re-derive it. If false, force matching on instantaneous forces would not yield the PMF.
  • standard math Atomistic and CG systems are in canonical equilibrium and the CG potential is the potential of mean force given by (2.4)
    Standard statistical mechanics; used in Sections 2.1 and Appendix A.
  • domain assumption ACE basis completeness and approximation theory: as body order, cutoff, and degree go to infinity, any symmetric function can be represented
    Cited from references 2, 3, 13; supports the 'systematically improvable' claim but is not proved in this paper.
  • domain assumption CG interactions are local within Rcut
    Explicitly flagged in Section 2.2: no theoretical justification; empirical support cited.
  • domain assumption Coarse-graining mapping is linear, species zeta_I is the multiset of atomic species, and momentum-space consistency conditions hold
    Assumed in Section 2.1; may fail for molecules where internal topology matters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Many-Body Coarse-Grained Molecular Dynamics with the Atomic Cluster Expansion." pith.science (2026). https://pith.science/paper/MWSYAK65

@misc{pith2026250204661,
  author       = {Pith},
  title        = {Pith review of: Many-Body Coarse-Grained Molecular Dynamics with the Atomic Cluster Expansion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MWSYAK65}},
  note         = {Machine review of arXiv:2502.04661}
}
read the original abstract

Molecular dynamics (MD) simulations provide detailed insight into atomic-scale mechanisms but are inherently restricted to small spatio-temporal scales. Coarse-grained molecular dynamics (CGMD) techniques allow simulations of much larger systems over extended timescales. In theory, these techniques can be quantitatively accurate, but common practice is to only target qualitatively correct behaviour of coarse-grained models. Recent advances in applying machine learning methodology in this setting are now being applied to create also quantitatively accurate CGMD models. We demonstrate how the Atomic Cluster Expansion parameterization (Drautz, 2019) can be used in this task to construct highly efficient, interpretable and accurate CGMD models. We focus in particular on exploring the role of many-body effects.

Figures

Figures reproduced from arXiv: 2502.04661 by the authors.

Figure 1
Figure 1. CG mapping for star polymer fluids in bulk [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Polymer fluids: RDFs generated by different CG models compared to the all-atom MD (black dots). CG Models ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) IBI (6) IBI (3) ERDF 0.058 0.152 4.711 0.183 0.845 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Polymer fluids: ADFs for different cutoff radii [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Structure of bulk methanol fluids. Left: the snapshot of MD trajec￾tory at t = 1.0 ns; Right: two levels of CG mappings for methanol. 3.3.2. Results. The ACE-CG potential U ACE(θ; R, ζ) is trained using the approach outlined in Section 2.2, employing training data acqu…
Figure 5
Figure 5. Figure 5: Methanol fluids: RDF for one-site description. 2 4 6 8 10 12 14 r (Å) 0.000 0.005 0.010 0.015 0.020 0.025 0.030 g(r) ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) MD (a) Carbon-carbon RDF. 2 4 6 8 10 r (Å) 0.000 0.005 0.010 0.015 0.020 0.025 g(r) ACE-CG (BO4) ACE-CG (BO3) ACE…
Figure 6
Figure 6. Figure 6: Methanol fluids: RDFs for the two-site description [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Methanol fluids: one-site ADF for different cutoff radii. 0.5 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 P( ) ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) MD (a) rcut = 4.5˚A 0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 P( ) ACE-CG (BO4) ACE-CG (B…
Figure 8
Figure 8. Figure 8: Methanol fluids: two-sites ADF for carbon group with different cutoff radii. Finally, [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Polymer fluids: ADF with rcut = 30˚A [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Methanol fluids: RDF for mixed carbon-oxygen group. 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.2 0.4 0.6 0.8 P( ) ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) MD (a) rcut = 4.5˚A 0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 P( ) ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) …
Figure 11
Figure 11. Figure 11: Methanol fluids: two-sites ADF for oxygen group with different cutoff radii. [5] A. Bart´ok, M. Payne, R. Kondor, and G. Cs´anyi. Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons. Phys. Rev. Lett., 104:136403, 2010. [6] I. Ba…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Graph-Coarsening for Machine Learning Coarse-grained Molecular Dynamics

    cond-mat.soft 2025-07 conditional novelty 4.0 of 10

    Graph-spectral coarsening (LVN/LVC) with MACE force matching yields coarse-grained models that match structural statistics of their own training trajectories for three small molecules.

  2. Fine-Tuning Universal Machine-Learned Interatomic Potentials: A Tutorial on Methods and Applications

    physics.comp-ph 2025-06 conditional novelty 4.0 of 10

    Fine-tuning universal MLIPs improves accuracy and data efficiency across electrolytes, defects, and interfaces, with some evidence of implicit long-range behavior that is not conclusive.

  3. A Study on the Fine-Tuning Performance of Universal Machine-Learned Interatomic Potentials (U-MLIPs)

    physics.comp-ph 2025-06 conditional novelty 4.0 of 10

    Fine-tuning universal MACE potentials on targeted datasets generally improves accuracy and convergence speed, though data selection, not the foundation model alone, determines success.

Reference graph

Works this paper leans on

75 extracted references · 74 canonical work pages · cited by 3 Pith papers

  1. [1]

    Ahmed and R

    A. Ahmed and R. Sadus. Phase diagram of the weeks-chandler-andersen potential from very low to high temperatures and pressures. Phys. Rev. E, 80(6):061101, 2009

  2. [2]

    Bachmayr, G

    M. Bachmayr, G. Csanyi, G. Dusson, R. Drautz, S. Etter, C. van der Oord, and C. Ortner. Atomic cluster expansion: Completeness, efficiency and stability. J. Comp. Phys., 454:110946, 2022

  3. [3]

    Bachmayr, G

    M. Bachmayr, G. Dusson, C. Ortner, and J. Thomas. Polynomial approximation of symmetric functions. Math. Comp., 93:811–839, 2024

  4. [4]

    Barth, K

    E. Barth, K. Kuczera, B. Leimkuhler, and R. Skeel. Algorithms for constrained molecular dynamics. J. Comput. Chem., 16(10):1192–1209, 1995. MANY-BODY COARSE-GRAINED MOLECULAR DYNAMICS WITH ACE 20 2 3 4 5 6 7 8 9 10 r (Å) 0.000 0.002 0.004 0.006 0.008 g(r) ACE-CG (BO4) ACE-CG (BO3) ACE-CG (BO2) MD Figure 10. Methanol fluids: RDF for mixed carbon-oxygen gro...

  5. [5]

    Bart´ ok, M

    A. Bart´ ok, M. Payne, R. Kondor, and G. Cs´ anyi. Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons. Phys. Rev. Lett., 104:136403, 2010

  6. [6]

    Batatia, D

    I. Batatia, D. Kovacs, G. Simm, C. Ortner, and G. Csanyi. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields. NeurIPS, 35:11423–11436, 2022

  7. [7]

    Behler and M

    J. Behler and M. Parrinello. Generalized neural-network representation of high-dimensional potential- energy surfaces. Phys. Rev. Lett., 98:146401, 2007

  8. [8]

    Bochkarev, Y

    A. Bochkarev, Y. Lysogorskiy, S. Menon, M. Qamar, M. Mrovec, and R. Drautz. Efficient parametrization of the atomic cluster expansion. Phys. Rev. Mater., 6:013804, Jan 2022

Show all 75 references
  1. [9]

    P. Bond, J. Holyoake, A. Ivetac, S. Khalid, and M. Sansom. Coarse-grained molecular dynamics simulations of membrane proteins and peptides. J. Struct. Biol., 157(3):593–605, 2007

  2. [10]

    Braun, C

    J. Braun, C. Ortner, Y. Wang, and L. Zhang. Higher order far-field boundary conditions for crystalline defects. arXiv preprint arXiv:2210.05573, 2022

  3. [11]

    H. Chen, C. Ortner, and Y. Wang. QM/MM methods for crystalline defects. part 3: machine-learned mm models. Multiscale Model. Simul., 20(4):1490–1518, 2022

  4. [12]

    Dynamic force matching: A method for constructing dynamical coarse-grained models with realistic time dependence

    Aram Davtyan, James F Dama, Gregory A Voth, and Hans C Andersen. Dynamic force matching: A method for constructing dynamical coarse-grained models with realistic time dependence. The Journal of Chemical Physics, 142(15), 2015

  5. [13]

    R. Drautz. Atomic cluster expansion for accurate and transferable interatomic potentials. Phys. Rev. B, 99:014104, 2019

  6. [14]

    Durumeric, Y

    A. Durumeric, Y. Chen, F. No´ e, and C. Clementi. Learning data efficient coarse-grained molecular dynamics from forces and noise. arXiv preprint arXiv:2407.01286, 2024. MANY-BODY COARSE-GRAINED MOLECULAR DYNAMICS WITH ACE 21

  7. [15]

    Duschatko, X

    B. Duschatko, X. Fu, C. Owen, Y. Xie, A. Musaelian, T. Aakkola, and B. Kozinsky. Thermodynamically informed multimodal learning of high-dimensional free energy models in molecular coarse graining. arXiv preprint arXiv:2405.19386, 2024

  8. [16]

    P. Ge, L. Zhang, and H. Lei. Machine learning assisted coarse-grained molecular dynamics modeling of meso-scale interfacial fluids. J. Chem. Phys., 158(6), 2023

  9. [17]

    Van Der Giessen, P

    E. Van Der Giessen, P. A. Schultz, N. Bertin, V. V. Bulatov, W. Cai, G. Cs´ anyi, S. M. Foiles, M. G. D. Geers, C. Gonz´ alez, M. H¨ utter, et al. Roadmap on multiscale materials modeling. Model. Simul. Mater. Sci. Eng., 28(4):043001, 2020

  10. [18]

    Constructing many-body dissipative particle dynamics models of fluids from bottom-up coarse-graining

    Yining Han, Jaehyeok Jin, and Gregory A Voth. Constructing many-body dissipative particle dynamics models of fluids from bottom-up coarse-graining. The Journal of Chemical Physics, 154(8), 2021

  11. [19]

    Hij´ on, P

    C. Hij´ on, P. Espa˜ nol, E. Vanden-Eijnden, and R. Delgado-Buscalioni. Mori–zwanzig formalism as a practical computational tool. Faraday Discuss., 144:301–322, 2010

  12. [20]

    A coarse-grained polymer model for studying the glass transition

    Hsiao-Ping Hsu and Kurt Kremer. A coarse-grained polymer model for studying the glass transition. The Journal of chemical physics, 150(9), 2019

  13. [21]

    B. E. Husic, N. E. Charron, D. Lemm, J. Wang, A. P´ erez, M. Majewski, A. Kr¨ amer, Y. Chen, S. Olsson, G. De Fabritiis, et al. Coarse graining molecular dynamics with graph neural networks. J. Chem. Phys., 153(19), 2020

  14. [22]

    A multiscale coarse-graining method for biomolecular systems

    Sergei Izvekov and Gregory A Voth. A multiscale coarse-graining method for biomolecular systems. The Journal of Physical Chemistry B, 109(7):2469–2473, 2005

  15. [23]

    Bottom-up coarse-graining: Principles and perspectives

    Jaehyeok Jin, Alexander J Pak, Aleksander EP Durumeric, Timothy D Loose, and Gregory A Voth. Bottom-up coarse-graining: Principles and perspectives. Journal of chemical theory and computation, 18(10):5759–5791, 2022

  16. [24]

    Temperature and phase transferable bottom-up coarse- grained models

    Jaehyeok Jin, Alvin Yu, and Gregory A Voth. Temperature and phase transferable bottom-up coarse- grained models. Journal of chemical theory and computation, 16(11):6823–6842, 2020

  17. [25]

    S. T. John and G. Csanyi. Many-body coarse-grained interactions using gaussian approximation potentials. J. Phys. Chem. B, 121(48):10934–10949, 2017

  18. [26]

    De Jong, G

    D. De Jong, G. Singh, D. Bennett, C. Arnarez, T. Wassenaar, L. Schafer, X. Periole, D. Tieleman, and S. Marrink. Improved parameters for the martini coarse-grained protein force field. J. Chem. Theory Com- put., 9(1):687–697, 2013

  19. [27]

    G. A. Kaminski, R. A. Friesner, J. Tirado-Rives, and W. L. Jorgensen. Evaluation and reparametrization of the opls-aa force field for proteins via comparison with accurate quantum chemical calculations on peptides. J. Phys. Chem. B, 105(28):6474–6487, 2001

  20. [28]

    Klippenstein, M

    V. Klippenstein, M. Tripathy, G. Jung, F. Schmid, and N. F. A. van der Vegt. Introducing memory in coarse-grained molecular simulations. J. Phys. Chem. B, 125(19):4931–4954, 2021

  21. [29]

    Kmiecik, D

    S. Kmiecik, D. Gront, M. Kolinski, L. Wieteska, A. Dawid, and A. Kolinski. Coarse-grained protein models and their applications. Chem. Rev., 116(14):7898–7936, 2016

  22. [30]

    Kohler, Y

    J. Kohler, Y. Chen, A. Kramer, C. Clementi, and F. No´ e. Flow-matching: Efficient coarse-graining of molecular dynamics without forces. J. Chem. Theory Comput., 19(3):942–952, 2023

  23. [31]

    D. P. Kov´ acs, C. van der Oord, J. Kucera, A. E. A. Allen, D. J. Cole, C. Ortner, and G. Cs´ anyi. Lin- ear atomic cluster expansion force fields for organic molecules: beyond rmse. J. Chem. Theory Comput., 17(12):7696–7711, 2021

  24. [32]

    A. H. Larsen, J. J. Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, et al. The atomic simulation environment—a python library for working with atoms. J. Phys.: Condens. Matter, 29(27):273002, 2017

  25. [33]

    H. Lei, N. A. Baker, and X. Li. Data-driven parameterization of the generalized langevin equation. Proc. Natl. Acad. Sci., 113(50):14183–14188, 2016

  26. [34]

    Lemke and C

    T. Lemke and C. Peter. Neural network based prediction of conformational free energies-a new route toward coarse-grained simulation models. J. Chem. Theory Comput., 13(12):6213–6221, 2017

  27. [35]

    Z. Li, X. Bian, X. Li, and G. E. Karniadakis. Incorporation of memory effects in coarse-grained modeling via the mori-zwanzig formalism. J. Chem. Phys., 143(24), 2015

  28. [36]

    Z. Li, H. S. Lee, E. Darve, and G. E. Karniadakis. Computing the non-markovian coarse-grained interactions derived from the mori–zwanzig formalism in molecular systems: Application to polymer melts. J. Chem. Phys., 146(1), 2017. MANY-BODY COARSE-GRAINED MOLECULAR DYNAMICS WITH ACE 22

  29. [37]

    Coarse-graining with equivari- ant neural networks: A path toward accurate and data-efficient models

    Timothy D Loose, Patrick G Sahrmann, Thomas S Qu, and Gregory A Voth. Coarse-graining with equivari- ant neural networks: A path toward accurate and data-efficient models. The Journal of Physical Chemistry B, 127(49):10564–10572, 2023

  30. [38]

    Lu and E

    J. Lu and E. Vanden-Eijnden. Exact dynamical coarse-graining without time-scale separation. J. Chem. Phys., 141(4):07B619 1, 2014

  31. [39]

    Lysogorskiy, C

    Y. Lysogorskiy, C. van der Oord, A. Bochkarev, S. Menon, M. Rinaldi, T. Hammerschmidt, M. Mrovec, A. Thompson, G. Csanyi, C. Ortner, et al. Performant implementation of the atomic cluster expansion (pace) and application to copper and silicon. Npj Comput. Mater., 7(1):1–12, 2021

  32. [40]

    Lyu and H

    L. Lyu and H. Lei. Construction of coarse-grained molecular dynamics with many-body non-markovian memory. arXiv preprint arXiv:2304.09044, 2023

  33. [41]

    Majewski, A

    M. Majewski, A. P´ erez, P. Th¨ olke, S. Doerr, N. E. Charron, T. Giorgino, B. E. Husic, C. Clementi, F. No´ e, and G. De Fabritiis. Machine learning coarse-grained potentials of protein thermodynamics.Nature Communications, 14(1):5739, 2023

  34. [42]

    S. J. Marrink and D. P. Tieleman. Perspective on the martini model. Chem. Soc. Rev., 42(16):6801–6822, 2013

  35. [43]

    Mones, N

    L. Mones, N. Bernstein, and G. Cs´ anyi. Exploration, sampling, and reconstruction of free energy surfaces with gaussian process regression. J. Chem. Theory Comput., 12(10):5100–5110, 2016

  36. [44]

    T. C. Moore, C. R. Iacovella, and C. McCabe. Derivation of coarse-grained potentials via multistate iterative boltzmann inversion. J. Chem. Phys., 140(22), 2014

  37. [45]

    No´ e, A

    F. No´ e, A. Tkatchenko, K.-R. M¨ uller, and C. Clementi. Machine learning for molecular simulation.Annu. Rev. Phys. Chem., 71:361–390, 2020

  38. [46]

    W. G. Noid. Perspective: Coarse-grained models for biomolecular systems. J. Chem. Phys. , 139(9):09B201 1, 2013

  39. [47]

    W. G. Noid, J.-W. Chu, G. S. Ayton, V. Krishna, S. Izvekov, G. A. Voth, A. Das, and H. C. Andersen. The multiscale coarse-graining method. i. a rigorous bridge between atomistic and coarse-grained models. J. Chem. Phys., 128(24):244114, 2008

  40. [48]

    Pagonabarraga and D

    I. Pagonabarraga and D. Frenkel. Dissipative particle dynamics for interacting systems. J. Chem. Phys., 115(11):5015–5026, 2001

  41. [49]

    T. K. Patra, T. D. Loeffler, H. Chan, M. J. Cherukara, B. Narayanan, and S. K. R. Sankaranarayanan. A coarse-grained deep neural network model for liquid water. Appl. Phys. Lett., 115(19), 2019

  42. [50]

    Pretti and M

    E. Pretti and M. S. Shell. A microcanonical approach to temperature-transferable coarse-grained models using the relative entropy. J. Chem. Phys., 155(9), 2021

  43. [51]

    Reith, M

    D. Reith, M. P¨ utz, and F. M¨ uller-Plathe. Deriving effective mesoscale potentials from atomistic simulations. J. Comput. Chem., 24(13):1624–1636, 2003

  44. [52]

    Dynamical properties across different coarse-grained models for ionic liquids

    Joseph F Rudzinski, Sebastian Kloth, Svenja W¨ orner, Tamisra Pal, Kurt Kremer, Tristan Bereau, and Michael Vogel. Dynamical properties across different coarse-grained models for ionic liquids. Journal of Physics: Condensed Matter, 33(22):224001, 2021

  45. [53]

    Sachs, W

    M. Sachs, W. G. Stark, R. J. Maurer, and C. Ortner. Equivariant Representation of Configuration- Dependent Friction Tensors in Langevin Heatbaths. ArXiv e-prints, 2407.13935, 2024

  46. [54]

    On the emergence of machine-learning methods in bottom-up coarse-graining

    Patrick G Sahrmann and Gregory A Voth. On the emergence of machine-learning methods in bottom-up coarse-graining. Current Opinion in Structural Biology, 90:102972, 2025

  47. [55]

    A. Shapeev. Moment tensor potentials: A class of systematically improvable interatomic potentials. Mul- tiscale Model. Simul., 14:1153–1173, 2016

  48. [56]

    Shireen, H

    Z. Shireen, H. Weeratunge, A. Menzel, A. W. Phillips, R. G. Larson, K. Smith-Miles, and E. Hajizadeh. A machine learning enabled hybrid optimization framework for efficient coarse-graining of a model polymer. npj Comput. Mater., 8(1):224, 2022

  49. [57]

    Sprik and G

    M. Sprik and G. Ciccotti. Free energy from constrained molecular dynamics. J. Chem. Phys., 109(18):7737– 7744, 1998

  50. [58]

    Stecher, N

    T. Stecher, N. Bernstein, and G. Cs´ anyi. Free energy surface reconstruction from umbrella samples using gaussian process regression. J. Chem. Theory Comput., 10(9):4079–4097, 2014

  51. [59]

    Stoltz, M

    G. Stoltz, M. Rousset, et al. Free energy computations: A mathematical perspective. World Scientific, 2010

  52. [60]

    Stukowski

    A. Stukowski. Visualization and analysis of atomistic simulation data with ovito–the open visualization tool. Model. Simul. Mater. Sci. Eng., 18(1):015012, 2009. MANY-BODY COARSE-GRAINED MOLECULAR DYNAMICS WITH ACE 23

  53. [61]

    Teza and A

    G. Teza and A. L. Stella. Exact coarse graining preserves entropy production out of equilibrium. Phys. Rev. Lett., 125(11):110601, 2020

  54. [62]

    Thaler, M

    S. Thaler, M. Stupp, and J. Zavadlav. Deep coarse-grained potentials via relative entropy minimization. J. Chem. Phys., 157(24):244103, 2022

  55. [63]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, et al. Lammps-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales.Comput. ...

  56. [64]

    G. A. Voth. Coarse-graining of condensed phase and biomolecular systems. CRC press, 2008

  57. [65]

    J. Wang, N. Charron, B. Husic, S. Olsson, F. No´ e, and C. Clementi. Multi-body effects in a coarse-grained protein force field. J. Chem. Phys., 154(16), 2021

  58. [66]

    J. Wang, S. Olsson, C. Wehmeyer, A. Perez, N. E. Charron, G. De Fabritiis, F. Noe, and C. Clementi. Machine learning of coarse-grained molecular dynamics force fields. ACS Cent. Sci., 5(5):755–767, 2019

  59. [67]

    S. Wang, Z. Ma, and W. Pan. Data-driven coarse-grained modeling of polymers in solution with structural and dynamic properties conserved. Soft Matter, 16(36):8330–8344, 2020

  60. [68]

    Y. Wang, H. Chen, M. Liao, C. Ortner, H. Wang, and L. Zhang. A posteriori error estimates for adaptive QM/MM coupling methods. SIAM J. Sci. Comput., 43(4):A2785–A2808, 2021

  61. [69]

    Y. Wang, W. G. Noid, P. Liu, and G. A. Voth. Effective force coarse-graining. Phys. Chem. Chem. Phys., 11(12):2002–2015, 2009

  62. [70]

    Y. Wang, S. Patel, and C. Ortner. A theoretical case study of the generalization of machine-learned potentials. Comput. Methods Appl. Mech. Engrg., 422:116831, 2024

  63. [71]

    W. C. Witt, C. van der Oord, E. Gel vzinyt˙ e, T. J¨ arvinen, A. Ross, J. P. Darby, C. H. Ho, W. J. Baldwin, M. Sachs, J. Kermode, N. Bernstein, G. Cs´ anyi, and C. Ortner. Acepotentials.jl: A julia implementation of the atomic cluster expansion. J. Chem. Phys., 159:164101, 2023

  64. [72]

    K-means clustering coarse-graining (kmc-cg): A next generation methodology for determining optimal coarse-grained mappings of large biomolecules

    Jiangbo Wu, Weizhi Xue, and Gregory A Voth. K-means clustering coarse-graining (kmc-cg): A next generation methodology for determining optimal coarse-grained mappings of large biomolecules. Journal of Chemical Theory and Computation, 19(23):8987–8997, 2023

  65. [73]

    Multibody terms in protein coarse-grained models: A top-down perspective

    Iryna Zaporozhets and Cecilia Clementi. Multibody terms in protein coarse-grained models: A top-down perspective. The Journal of Physical Chemistry B, 127(31):6920–6927, 2023

  66. [74]

    Zavadlav, S

    J. Zavadlav, S. J. Marrink, and M. Praprotnik. Multiscale simulation of protein hydration using the swinger dynamical clustering algorithm. J. Chem. Theory Comput., 14(3):1754–1761, 2018

  67. [75]

    Zhang, J

    L. Zhang, J. Han, H. Wang, R. Car, et al. Deepcg: Constructing coarse-grained models via deep neural networks. J. Chem. Phys., 149(3), 2018. Yangshuai W ang, Department of Mathematics, National University of Singapore, 10 Lower Kent Ridge Road, Singapore Email address: yswang@...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.