pith. sign in

arxiv: 2506.12893 · v4 · submitted 2025-06-15 · ⚛️ physics.comp-ph

Analytical coarse grained potential parameterization by Reinforcement Learning for anisotropic cellulose

Xu Dong This is my paper

Pith reviewed 2026-05-19 09:35 UTC · model grok-4.3

classification ⚛️ physics.comp-ph
keywords coarse-grained modelingreinforcement learningcellulose nanocrystalsanisotropyBoltzmann inversionmechanical propertiesmolecular dynamics
0
0 comments X

The pith

Reinforcement learning parameterizes an analytical coarse-grained potential for cellulose that generalizes to mechanical properties outside the training set.

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

The paper presents a bottom-up method that uses reinforcement learning together with Boltzmann inversion to build an analytical coarse-grained model of cellulose nanocrystals. The model targets the anisotropic structure that arises from hydrogen bonding and the associated polymer stiffness. A central result is that the trained potential reproduces dynamic mechanical responses under conditions not encountered during training and without any further adjustment. This outcome indicates that reinforcement learning can generate coarse-grained potentials that stay physically interpretable while remaining accurate for mesoscale simulations of cellulose materials.

Core claim

An analytical coarse-grained potential for cellulose is parameterized directly by reinforcement learning in an extended bottom-up scheme that incorporates Boltzmann inversion. The resulting model captures both the anisotropy and the stiffness of cellulose nanocrystals. It reproduces dynamic mechanical properties under circumstances different from those used in training and without requiring additional training, thereby showing that reinforcement learning can produce a coarse-grained potential that is both physically explainable and powerful.

What carries the argument

Reinforcement learning parameterization of an analytical coarse-grained potential combined with Boltzmann inversion, used to encode cellulose anisotropy and polymer stiffness.

If this is right

  • The model supports mesoscopic simulations of cellulose material development and manufacture where atomistic methods remain too costly.
  • Dynamic mechanical properties can be obtained for conditions outside the original training set without retraining.
  • Reinforcement learning supplies a route to coarse-grained potentials that remain both interpretable and accurate for anisotropic polymers.
  • The approach extends the bottom-up parameterization workflow to systems whose structure is dominated by directional hydrogen bonds.

Where Pith is reading between the lines

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

  • The same RL-plus-inversion workflow may transfer to other hydrogen-bonded anisotropic nanomaterials such as chitin or silk fibrils.
  • Once the analytical form is fixed, the learned parameters could be reused across related cellulose polymorphs with only minor recalibration.
  • Embedding the potential in larger-scale continuum models would allow direct prediction of macroscopic composite behavior from the mesoscale anisotropy.

Load-bearing premise

Reinforcement learning combined with Boltzmann inversion can produce an analytical coarse-grained potential that is physically explainable and generalizes to mechanical properties outside the training set for representing cellulose anisotropy.

What would settle it

A direct comparison in which the coarse-grained model is applied to a shear or tensile test at an angle or strain rate absent from the training data and the predicted stiffness or relaxation behavior deviates markedly from corresponding atomistic reference simulations.

Figures

Figures reproduced from arXiv: 2506.12893 by Xu Dong.

Figure 1
Figure 1. Figure 1: Orthotropic transverse section and characteristi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mapping and topology. (a) Mapping and (b) mapped ov [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: RL diagrams. (a) Standard RL diagram. In the standa [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training procedures. The BD properties (axial ela [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Training convergence and statistics-guided redu [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Axial stretching was performed in an xyz-periodic cell wit [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: Bonded properties. (a) Axial elastic modulus. The [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Nonbonded properties. (a) Behaviors of CG models i [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Draw-out and Tear-apart. (a) Draw-out and (b) curv [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Adhesion and Bending in aperiodic systems. (a) Adh [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Transverse arrangement. (a) Transverse arrange [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Brick-and-mortar. (a) Brick-and-mortar models [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Computation efficiency data (a) and (b) relative effi [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
read the original abstract

Cellulose nanocrystals (CNCs) are a type of cellulose with excellent mechanical performance and other merit attributes. According to previous reports, hydrogen bonds play a pivotal role in the anisotropic structure of the CNC. Understanding the structure and mechanical behavior of CNC on a mesoscopic scale is critical for the development and manufacture of cellulose materials. However, experimental observations and atomistic simulations are not appropriate on the mesoscopic scale. In this study, we introduce an analytical coarse-grained (CG) potential following an extended bottom-up approach that is directly parameterized using Reinforcement Learning (RL). RL is a powerful tool for industrial and academic applications in various fields. Nevertheless, the potential of RL has not yet been fully exploited in the field of molecular dynamics. The RL and Boltzmann inversion methods were employed to develop a novel CG model of cellulose to represent its anisotropy and polymer stiffness. The resultant CG model is not limited to the target properties for training, and can reproduce the dynamics mechanical properties under other circumstances without additional training. This model confirms that RL can construct a CG potential that is both physically explainable and powerful.

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 / 1 minor

Summary. The manuscript introduces an analytical coarse-grained (CG) potential for cellulose nanocrystals, parameterized via Reinforcement Learning (RL) combined with Boltzmann inversion in an extended bottom-up approach. It claims this model captures anisotropy arising from hydrogen bonds, generalizes to reproduce dynamic mechanical properties under untrained conditions without retraining, and yields a physically explainable potential.

Significance. If the transferability to out-of-distribution mechanical properties and physical explainability are quantitatively validated, the work could advance RL-based parameterization of analytical CG potentials for anisotropic biomaterials, offering a scalable alternative to purely data-driven or tabulated CG models for mesoscale cellulose simulations.

major comments (2)
  1. [Abstract] Abstract: The central claim that 'the resultant CG model is not limited to the target properties for training, and can reproduce the dynamics mechanical properties under other circumstances without additional training' is load-bearing for the contribution but is unsupported by any quantitative error metrics, comparison baselines to atomistic references, description of the reward function, or specification of the 'other circumstances' tested.
  2. [Methods] Methods/Results (RL parameterization section): The assertion of physical explainability for the analytical CG potential requires explicit demonstration (e.g., via parameter interpretation linking to hydrogen-bond physics or sensitivity analysis); without this, the RL optimization risks being a black-box fit to training observables rather than encoding transferable anisotropic physics.
minor comments (1)
  1. [Abstract] Abstract: 'dynamics mechanical properties' should read 'dynamic mechanical properties'; 'merit attributes' is unclear and could be rephrased for precision.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation of our results on RL-parameterized analytical CG potentials for anisotropic cellulose. We address each major point below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 'the resultant CG model is not limited to the target properties for training, and can reproduce the dynamics mechanical properties under other circumstances without additional training' is load-bearing for the contribution but is unsupported by any quantitative error metrics, comparison baselines to atomistic references, description of the reward function, or specification of the 'other circumstances' tested.

    Authors: We agree that the abstract would be strengthened by including quantitative details. The full manuscript reports direct comparisons to atomistic reference simulations for mechanical properties (Young's moduli and stress-strain response) under conditions outside the training set, such as varying strain rates and temperatures, with root-mean-square errors provided in the Results section. The reward function combines structural matching via Boltzmann inversion with mechanical property targets and is specified in the Methods. In revision we will condense these elements into the abstract, explicitly naming the out-of-distribution tests (e.g., shear and tensile loading at untrained temperatures). revision: yes

  2. Referee: [Methods] Methods/Results (RL parameterization section): The assertion of physical explainability for the analytical CG potential requires explicit demonstration (e.g., via parameter interpretation linking to hydrogen-bond physics or sensitivity analysis); without this, the RL optimization risks being a black-box fit to training observables rather than encoding transferable anisotropic physics.

    Authors: We acknowledge the value of an explicit demonstration. The analytical functional form was deliberately chosen to separate isotropic and anisotropic terms that reflect the directional hydrogen-bond network in cellulose; the RL procedure optimizes coefficients within this physically motivated ansatz rather than learning an arbitrary mapping. To make this transparent we will add a dedicated subsection that (i) interprets the magnitude of the anisotropic coefficients in terms of known hydrogen-bond strengths and (ii) presents a one-at-a-time sensitivity analysis showing how each term affects the predicted anisotropy. These additions will be placed in the revised Methods/Results. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation uses standard RL parameterization with empirical generalization claims

full rationale

The paper applies reinforcement learning together with Boltzmann inversion to fit an analytical CG potential to target structural and stiffness properties of cellulose. The central claim of reproducing dynamic mechanical properties outside the training set is presented as an empirical outcome of the optimized potential rather than a mathematical identity or self-referential fit. No equations are shown that reduce the out-of-distribution predictions to the training observables by construction, and no self-citation chain is invoked to justify uniqueness or the functional form. The method is a data-driven parameterization technique whose transferability assertions, if supported by separate validation, remain independent of the fitting procedure itself. This is the expected non-circular outcome for a parameterization study that does not redefine its targets as predictions.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the ledger captures the stated domain assumption about hydrogen bonds and the ad-hoc assumption that RL yields explainable potentials. No explicit free parameters or invented entities are described.

free parameters (1)
  • Analytical CG potential parameters
    Optimized by RL to reproduce target anisotropy and stiffness properties.
axioms (1)
  • domain assumption Hydrogen bonds play a pivotal role in the anisotropic structure of the CNC.
    Cited from previous reports in the abstract.

pith-pipeline@v0.9.0 · 5709 in / 1202 out tokens · 64227 ms · 2026-05-19T09:35:03.495284+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages

  1. [1]

    O verview of cellulose nanomaterials, their capabilities and applications

    Robert J Moon, Gregory T Schueneman, and John Simonsen. O verview of cellulose nanomaterials, their capabilities and applications. Jom, 68:2383–2394, 2016

    work page 2016

  2. [2]

    Review of nanocellulose for sustaina ble future materials

    Joo-Hyung Kim, Bong Sup Shim, Heung Soo Kim, Young-Jun Le e, Seung-Ki Min, Daseul Jang, Zafar Abas, and Jaehwan Kim. Review of nanocellulose for sustaina ble future materials. International Journal of Precision Engineering and Manufacturing-Green Technology , 2:197–213, 2015

    work page 2015

  3. [3]

    Towards an eco-fri endly deconstruction of agro-industrial biomass and preparation of renewable cellulose nanomaterials: A re view

    Hwee Li Teo and Roswanira Abdul Wahab. Towards an eco-fri endly deconstruction of agro-industrial biomass and preparation of renewable cellulose nanomaterials: A re view. International Journal of Biological Macro- molecules, 161:1414–1430, 2020

    work page 2020

  4. [4]

    Cellulose nanomaterials review: structure, properties and nanocomposites

    Robert J Moon, Ashlie Martini, John Nairn, John Simonsen , and Jeff Youngblood. Cellulose nanomaterials review: structure, properties and nanocomposites. Chemical Society Reviews , 40(7):3941–3994, 2011

    work page 2011

  5. [5]

    Stimuli-responsive cellulose nanomateria ls for smart applications

    Qianqian Zhu, Simeng Liu, Jianzhong Sun, Jun Liu, C Josep h Kirubaharan, Honglei Chen, Weihua Xu, and Qianqian Wang. Stimuli-responsive cellulose nanomateria ls for smart applications. Carbohydrate Polymers, 235:115933, 2020

    work page 2020

  6. [6]

    Cellulose nanocrystals and cellulose nanofibrils based hydrogels for biomedical appli cations

    Haishun Du, Wei Liu, Miaomiao Zhang, Chuanling Si, Xinyu Zhang, and Bin Li. Cellulose nanocrystals and cellulose nanofibrils based hydrogels for biomedical appli cations. Carbohydrate polymers, 209:130–144, 2019

    work page 2019

  7. [7]

    W ater treatment via non-membrane inorganic nanoparticles/cellulose composites

    Jiwoo Yu, Aurelia C Wang, Mingyue Zhang, and Zhiqun Lin. W ater treatment via non-membrane inorganic nanoparticles/cellulose composites. Materials Today, 50:329–357, 2021

    work page 2021

  8. [8]

    Cellulose based composite foams and aerogels for advanced e nergy storage devices

    Huayu Liu, Haishun Du, Ting Zheng, Kun Liu, Xingxiang Ji, Ting Xu, Xinyu Zhang, and Chuanling Si. Cellulose based composite foams and aerogels for advanced e nergy storage devices. Chemical Engineering Journal, 426:130817, 2021

    work page 2021

  9. [9]

    Flexible and transparent cellulose-based ionic film as a humidity sensor

    Yang Wang, Lina Zhang, Jinping Zhou, and Ang Lu. Flexible and transparent cellulose-based ionic film as a humidity sensor. ACS Applied Materials & Interfaces , 12(6):7631–7638, 2020

    work page 2020

  10. [10]

    Dft study of the application of polymers cellulose and cellulose acetate for adsorption of metal ions (cd 2+, cu 2+ and cr 3+) potentially toxic

    Davi Texeira Reis, Igor Hernandes Santos Ribeiro, and D ouglas Henrique Pereira. Dft study of the application of polymers cellulose and cellulose acetate for adsorption of metal ions (cd 2+, cu 2+ and cr 3+) potentially toxic. Polymer Bulletin , 77:3443–3456, 2020

    work page 2020

  11. [11]

    Evaluati on of nanocellulose interaction with water pollutants using nanocellulose colloidal probes and molec ular dynamic simulations

    Chuantao Zhu, Susanna Monti, and Aji P Mathew. Evaluati on of nanocellulose interaction with water pollutants using nanocellulose colloidal probes and molec ular dynamic simulations. Carbohydrate polymers, 229:115510, 2020

    work page 2020

  12. [12]

    Linear, nonlinear and plastic bending deformation of cellulose nanocrystals

    Pan Chen, Yu Ogawa, Yoshiharu Nishiyama, Ahmed E Ismail , and Karim Mazeau. Linear, nonlinear and plastic bending deformation of cellulose nanocrystals. Physical Chemistry Chemical Physics , 18(29):19880– 19887, 2016

    work page 2016

  13. [13]

    Anisotropic elastic properties of cellulose measured using inelastic x-ray scattering

    Imke Diddens, Bridget Murphy, Michael Krisch, and Mart in Muller. Anisotropic elastic properties of cellulose measured using inelastic x-ray scattering. Macromolecules, 41(24):9755–9759, 2008

    work page 2008

  14. [14]

    Elastic modulus of single cellulose microfibrils from tunicate measured by atomic force microsc opy

    Shinichiro Iwamoto, Weihua Kai, Akira Isogai, and Tada hisa Iwata. Elastic modulus of single cellulose microfibrils from tunicate measured by atomic force microsc opy. Biomacromolecules, 10(9):2571–2576, 2009

    work page 2009

  15. [15]

    Mechanica l properties of cellulose nanomaterials studied by contact resonance atomic force microscopy

    Ryan Wagner, Robert J Moon, and Arvind Raman. Mechanica l properties of cellulose nanomaterials studied by contact resonance atomic force microscopy. Cellulose, 23:1031–1041, 2016

    work page 2016

  16. [16]

    Anisotropy of the elastic properties of crystalline cellulose i β from first principles density functional theory with van der waals interactions

    Fernando L Dri, Louis G Hector, Robert J Moon, and Pablo D Zavattieri. Anisotropy of the elastic properties of crystalline cellulose i β from first principles density functional theory with van der waals interactions. Cellulose, 20:2703–2718, 2013. 22

    work page 2013

  17. [17]

    Tensile str ength of i β crystalline cellulose predicted by molecular dynamics simulation

    Xiawa Wu, Robert J Moon, and Ashlie Martini. Tensile str ength of i β crystalline cellulose predicted by molecular dynamics simulation. Cellulose, 21:2233–2245, 2014

    work page 2014

  18. [18]

    Effects of nanofiber orientations on the fracture toughness of cellulose nanopaper

    Qinghua Meng, Bo Li, Teng Li, and Xi-Qiao Feng. Effects of nanofiber orientations on the fracture toughness of cellulose nanopaper. Engineering Fracture Mechanics, 194:350–361, 2018

    work page 2018

  19. [19]

    A multisc ale crack-bridging model of cellulose nanopaper

    Qinghua Meng, Bo Li, Teng Li, and Xi-Qiao Feng. A multisc ale crack-bridging model of cellulose nanopaper. Journal of the Mechanics and Physics of Solids , 103:22–39, 2017

    work page 2017

  20. [20]

    The stability of cellulos e: a statistical perspective from a coarse-grained model of hydrogen-bond networks

    Tongye Shen and S Gnanakaran. The stability of cellulos e: a statistical perspective from a coarse-grained model of hydrogen-bond networks. Biophysical journal , 96(8):3032–3040, 2009

    work page 2009

  21. [21]

    Role of hydrogen bonding in cellulose deformation: the leverage effect analy zed by molecular modeling

    Cyrus Djahedi, Malin Bergenstr ˚ ahle-Wohlert, Lars A B erglund, and Jakob Wohlert. Role of hydrogen bonding in cellulose deformation: the leverage effect analy zed by molecular modeling. Cellulose, 23:2315– 2323, 2016

    work page 2016

  22. [22]

    Cellulose and the role of hydrogen bonds: not in charge of everything

    Malin Wohlert, Tobias Benselfelt, Lars W ˚ agberg, Istv´ an Fur´ o, Lars A Berglund, and Jakob Wohlert. Cellulose and the role of hydrogen bonds: not in charge of everything. Cellulose, pages 1–23, 2022

    work page 2022

  23. [23]

    Atomistic s imulation of frictional sliding between cellulose iβ nanocrystals

    Xiawa Wu, Robert J Moon, and Ashlie Martini. Atomistic s imulation of frictional sliding between cellulose iβ nanocrystals. Tribology Letters, 52:395–405, 2013

    work page 2013

  24. [24]

    Hydrogen bonds dominated frictional stick-slip of cellulose nanocrystals

    Chi Zhang, Sinan Keten, Dominique Derome, and Jan Carme liet. Hydrogen bonds dominated frictional stick-slip of cellulose nanocrystals. Carbohydrate Polymers, 258:117682, 2021

    work page 2021

  25. [25]

    A revi ew of cellulose coarse-grained models and their applications

    Aleksandar Y Mehandzhiyski and Igor Zozoulenko. A revi ew of cellulose coarse-grained models and their applications. Polysaccharides, 2(2):257–270, 2021

    work page 2021

  26. [26]

    New patchy particle model with anisotropic patches for molecul ar dynamics simulations: Application to a coarse- grained model of cellulose nanocrystal

    Nicolas Rolland, Aleksandar Y Mehandzhiyski, Mohit Ga rg, Mathieu Linares, and Igor V Zozoulenko. New patchy particle model with anisotropic patches for molecul ar dynamics simulations: Application to a coarse- grained model of cellulose nanocrystal. Journal of Chemical Theory and Computation , 16(6):3699–3711, 2020

    work page 2020

  27. [27]

    A novel supra coarse-grained model for cellulo se

    Aleksandar Y Mehandzhiyski, Nicolas Rolland, Mohit Ga rg, Jakob Wohlert, Mathieu Linares, and Igor Zozoulenko. A novel supra coarse-grained model for cellulo se. Cellulose, 27:4221–4234, 2020

    work page 2020

  28. [28]

    Coarse-grained modeling of n anocellulose network towards understanding the mechanical performance

    Zhaofan Li and Wenjie Xia. Coarse-grained modeling of n anocellulose network towards understanding the mechanical performance. Extreme Mechanics Letters , 40:100942, 2020

    work page 2020

  29. [29]

    Opti mizing the mechanical properties of cellulose nanopaper through surface energy and critical length scale considerations

    Xin Qin, Shizhe Feng, Zhaoxu Meng, and Sinan Keten. Opti mizing the mechanical properties of cellulose nanopaper through surface energy and critical length scale considerations. Cellulose, 24:3289–3299, 2017

    work page 2017

  30. [30]

    Mechanics of c ellulose nanopaper using a scalable coarse- grained modeling scheme

    Upamanyu Ray, Zhenqian Pang, and Teng Li. Mechanics of c ellulose nanopaper using a scalable coarse- grained modeling scheme. Cellulose, 28(6):3359–3372, 2021

    work page 2021

  31. [31]

    Martini coarse-graine d model for crystalline cellulose microfibers

    Cesar A Lopez, Giovanni Bellesia, Antonio Redondo, Pau l Langan, Shishir PS Chundawat, Bruce E Dale, Siewert J Marrink, and S Gnanakaran. Martini coarse-graine d model for crystalline cellulose microfibers. The Journal of Physical Chemistry B , 119(2):465–473, 2015

    work page 2015

  32. [32]

    T he energy landscape for the interaction of the family 1 carbohydrate-binding module and the cellulo se surface is altered by hydrolyzed glycosidic bonds

    Lintao Bu, Gregg T Beckham, Michael F Crowley, Christop her H Chang, James F Matthews, Yannick J Bomble, William S Adney, Michael E Himmel, and Mark R Nimlos. T he energy landscape for the interaction of the family 1 carbohydrate-binding module and the cellulo se surface is altered by hydrolyzed glycosidic bonds. The Journal of Physical Chemistry B , 113(...

    work page 2009

  33. [33]

    A coarse-grained mod el for molecular dynamics simulations of native cellulose

    Jakob Wohlert and Lars A Berglund. A coarse-grained mod el for molecular dynamics simulations of native cellulose. Journal of Chemical Theory and Computation , 7(3):753–760, 2011

    work page 2011

  34. [34]

    Coarse-grain m odel for natural cellulose fibrils in explicit water

    Srinivas Goundla, Cheng Xiaolin, et al. Coarse-grain m odel for natural cellulose fibrils in explicit water. 2014

    work page 2014

  35. [35]

    Systematic bottom-up molecular coarse-graining via force and torque matching using anisotropic particles

    Huong TL Nguyen and David M Huang. Systematic bottom-up molecular coarse-graining via force and torque matching using anisotropic particles. The Journal of Chemical Physics , 156(18), 2022

    work page 2022

  36. [36]

    Coarse-grained simula tion of cellulose i β with application to long fibrils

    Bingxin Fan and Janna K Maranas. Coarse-grained simula tion of cellulose i β with application to long fibrils. Cellulose, 22:31–44, 2015. 23

    work page 2015

  37. [37]

    Mechanical p roperties of cellulose nanocrystal (cnc) bundles: Coarse-grained molecular dynamic simulation

    Majid G Ramezani and Behnoush Golchinfar. Mechanical p roperties of cellulose nanocrystal (cnc) bundles: Coarse-grained molecular dynamic simulation. Journal of Composites Science , 3(2):57, 2019

    work page 2019

  38. [38]

    Effects of inte rface properties on the mechanical properties of bio-inspired cellulose nanocrystal (cnc)-based material s

    Mehdi Shishehbor and Pablo D Zavattieri. Effects of inte rface properties on the mechanical properties of bio-inspired cellulose nanocrystal (cnc)-based material s. Journal of the Mechanics and Physics of Solids , 124:871–896, 2019

    work page 2019

  39. [39]

    Influence of alignment and microstructure featur es on the mechanical properties and failure mech- anisms of cellulose nanocrystals (cnc) films

    Mehdi Shishehbor, Hyeyoung Son, Md Nuruddin, Jeffrey P Y oungblood, Chelsea Davis, and Pablo D Za- vattieri. Influence of alignment and microstructure featur es on the mechanical properties and failure mech- anisms of cellulose nanocrystals (cnc) films. Journal of the Mechanical Behavior of Biomedical Materials , 118:104399, 2021

    work page 2021

  40. [40]

    Machine learning for molec- ular simulation

    Frank No´ e, Alexandre Tkatchenko, Klaus-Robert M¨ uller, and Cecilia Clementi. Machine learning for molec- ular simulation. Annual Review of Physical Chemistry , 71(1):361–390, 2020

    work page 2020

  41. [41]

    An overview about neural networks potentials in molecular dyn amics simulation

    Raidel Martin-Barrios, Edisel Navas-Conyedo, Xuyi Zh ang, Yunwei Chen, and Jorge Gul ´ ın-Gonz´ alez. An overview about neural networks potentials in molecular dyn amics simulation. International Journal of Quantum Chemistry , 124(11):e27389, 2024

    work page 2024

  42. [42]

    Explaining deep neural networks and beyond: A revi ew of methods and applications

    Wojciech Samek, Gr´ egoire Montavon, Sebastian Lapusc hkin, Christopher J Anders, and Klaus-Robert M¨ uller. Explaining deep neural networks and beyond: A revi ew of methods and applications. Proceed- ings of the IEEE , 109(3):247–278, 2021

    work page 2021

  43. [43]

    Machin e learning for multi-fidelity scale bridging and dynamical simulations of materials

    Rohit Batra and Subramanian Sankaranarayanan. Machin e learning for multi-fidelity scale bridging and dynamical simulations of materials. Journal of Physics: Materials , 3(3):031002, 2020

    work page 2020

  44. [44]

    Machine learning directed optimization of classical molecular mod eling force fields

    Bridgette J Befort, Ryan S DeFever, Garrett M Tow, Alexa nder W Dowling, and Edward J Maginn. Machine learning directed optimization of classical molecular mod eling force fields. Journal of Chemical Information and Modeling , 61(9):4400–4414, 2021

    work page 2021

  45. [45]

    Coarse-grained force field calibration based on multiobjective bayesian optimizatio n to simulate water diffusion in poly- ε-caprolactone

    Jesse M Sestito, Mary L Thatcher, Leshi Shu, Tequila AL H arris, and Yan Wang. Coarse-grained force field calibration based on multiobjective bayesian optimizatio n to simulate water diffusion in poly- ε-caprolactone. The Journal of Physical Chemistry A , 124(24):5042–5052, 2020

    work page 2020

  46. [46]

    Machine-learning assisted coarse-grained model for epoxies over wide ranges of temperatures and cross-linking degrees

    Ke Duan, Yonglyu He, Yijun Li, Jiahuan Liu, Junjie Zhang , Yujin Hu, Rongming Lin, Xuelin Wang, Weiming Deng, and Li Li. Machine-learning assisted coarse-grained model for epoxies over wide ranges of temperatures and cross-linking degrees. Materials and Design , 183, 2019

    work page 2019

  47. [47]

    Machine learning co arse grained models for water

    Henry Chan, Mathew J Cherukara, Badri Narayanan, Troy D Loeffler, Chris Benmore, Stephen K Gray, and Subramanian KRS Sankaranarayanan. Machine learning co arse grained models for water. Nature communications, 10(1):379, 2019

    work page 2019

  48. [48]

    Force-field coe fficient optimization of coarse-grained molecular dynamics models with a small computational budget

    M Razi, A Narayan, RM Kirby, and D Bedrov. Force-field coe fficient optimization of coarse-grained molecular dynamics models with a small computational budget. Computational Materials Science , 176:109518, 2020

    work page 2020

  49. [49]

    Reinforcement learning base d hybrid bond-order coarse-grained interatomic potentials for exploring mesoscale aggregation in liquid– liquid mixtures

    Anirban Chandra, Troy Loeffler, Henry Chan, Xiaoyu Wang, GB Stephenson, Michael J Servis, and Subra- manian KRS Sankaranarayanan. Reinforcement learning base d hybrid bond-order coarse-grained interatomic potentials for exploring mesoscale aggregation in liquid– liquid mixtures. The Journal of Chemical Physics , 159(2), 2023

    work page 2023

  50. [50]

    Rlcg: When reinforcement learning meets coarse graining

    Shenghao Wu, Tianyi Liu, Zhirui Wang, Wen Yan, and Yingx iang Yang. Rlcg: When reinforcement learning meets coarse graining. In NeurIPS 2022 AI for Science: Progress and Promises , 2022

    work page 2022

  51. [51]

    L earning in continuous action space for developing high dimensional potential energy models

    Sukriti Manna, Troy D Loeffler, Rohit Batra, Suvo Banik, H enry Chan, Bilvin Varughese, Kiran Sasikumar, Michael Sternberg, Tom Peterka, Mathew J Cherukara, et al. L earning in continuous action space for developing high dimensional potential energy models. Nature communications, 13(1):368, 2022

    work page 2022

  52. [52]

    A tour of reinforcement learning: The v iew from continuous control

    Benjamin Recht. A tour of reinforcement learning: The v iew from continuous control. Annual Review of Control, Robotics, and Autonomous Systems , 2(1):253–279, 2019

    work page 2019

  53. [53]

    From predicting to decision making: Reinforcement learning in biomedicine

    Xuhan Liu, Jun Zhang, Zhonghuai Hou, Yi Isaac Yang, and Y i Qin Gao. From predicting to decision making: Reinforcement learning in biomedicine. Wiley Interdisciplinary Reviews: Computational Molecular S cience, 14(4):e1723, 2024. 24

    work page 2024

  54. [54]

    Lammps-a flexible simulation tool for particle-based materials modeling at t he atomic, meso, and continuum scales

    Aidan P Thompson, H Metin Aktulga, Richard Berger, Dan S Bolintineanu, W Michael Brown, Paul S Crozier, Pieter J In’t Veld, Axel Kohlmeyer, Stan G Moore, Tr ung Dac Nguyen, et al. Lammps-a flexible simulation tool for particle-based materials modeling at t he atomic, meso, and continuum scales. Computer physics communications, 271:108171, 2022

    work page 2022

  55. [55]

    Dr eiding: a generic force field for molecular simulations

    Stephen L Mayo, Barry D Olafson, and William A Goddard. Dr eiding: a generic force field for molecular simulations. Journal of Physical chemistry , 94(26):8897–8909, 1990

    work page 1990

  56. [56]

    Mach ine learning potentials with the iterative boltzmann inversion: training to experiment

    Sakib Matin, Alice EA Allen, Justin Smith, Nicholas Lub bers, Ryan B Jadrich, Richard Messerly, Benjamin Nebgen, Ying Wai Li, Sergei Tretiak, and Kipton Barros. Mach ine learning potentials with the iterative boltzmann inversion: training to experiment. Journal of Chemical Theory and Computation , 20(3):1274– 1281, 2024

    work page 2024

  57. [57]

    δ-learning applied to coarse-grained homogeneous liquids

    Aditi Khot and Brett M Savoie. δ-learning applied to coarse-grained homogeneous liquids. The Journal of Chemical Physics , 159(5), 2023

    work page 2023

  58. [58]

    Learning pair poten- tials using differentiable simulations

    Wujie Wang, Zhenghao Wu, Johannes CB Dietschreit, and R afael G´ omez-Bombarelli. Learning pair poten- tials using differentiable simulations. The Journal of Chemical Physics , 158(4), 2023

    work page 2023

  59. [59]

    Flap opening dynamics in hiv-1 protease explored with a coarse-grained model

    Valentina Tozzini, Joanna Trylska, Chia-en Chang, and J Andrew McCammon. Flap opening dynamics in hiv-1 protease explored with a coarse-grained model. Journal of structural biology , 157(3):606–615, 2007

    work page 2007

  60. [60]

    Versa- tile object-oriented toolkit for coarse-graining applica tions

    Victor Ruhle, Christoph Junghans, Alexander Lukyanov , Kurt Kremer, and Denis Andrienko. Versa- tile object-oriented toolkit for coarse-graining applica tions. Journal of chemical theory and computation , 5(12):3211–3223, 2009

    work page 2009

  61. [61]

    Direct shape optimization through deep reinforcement lear ning

    Jonathan Viquerat, Jean Rabault, Alexander Kuhnle, Ha ssan Ghraieb, Aur´ elien Larcher, and Elie Hachem. Direct shape optimization through deep reinforcement lear ning. Journal of Computational Physics , 428:110080, 2021

    work page 2021

  62. [62]

    Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic acto r

    Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Serge y Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic acto r. In International conference on machine learning , pages 1861–1870. Pmlr, 2018

    work page 2018

  63. [63]

    Stable-baselines3: Reliable reinforcement learning impl ementations

    Antonin Raffin, Ashley Hill, Adam Gleave, Anssi Kanervis to, Maximilian Ernestus, and Noah Dormann. Stable-baselines3: Reliable reinforcement learning impl ementations. Journal of Machine Learning Research , 22(268):1–8, 2021

    work page 2021

  64. [64]

    What makes a reward model a good teacher? an optimization perspective

    Noam Razin, Zixuan Wang, Hubert Strauss, Stanley Wei, J ason D Lee, and Sanjeev Arora. What makes a reward model a good teacher? an optimization perspective. arXiv preprint arXiv:2503.15477 , 2025

    work page arXiv 2025

  65. [65]

    Gromacs: High performance molecular simulations through multi-level parallelism from laptops to supercomputers

    Mark James Abraham, Teemu Murtola, Roland Schulz, Szil ´ ard P´ all, Jeremy C Smith, Berk Hess, and Erik Lindahl. Gromacs: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX, 1:19–25, 2015

    work page 2015

  66. [66]

    Implementation of the charmm force field in gromacs: analysis of protein stability effects from correction maps, virtual interaction sites, and water models

    Par Bjelkmar, Per Larsson, Michel A Cuendet, Berk Hess, and Erik Lindahl. Implementation of the charmm force field in gromacs: analysis of protein stability effects from correction maps, virtual interaction sites, and water models. Journal of chemical theory and computation , 6(2):459–466, 2010

    work page 2010

  67. [67]

    Charmm36 all-a tom additive protein force field: Validation based on comparison to nmr data

    Jing Huang and Alexander D MacKerell Jr. Charmm36 all-a tom additive protein force field: Validation based on comparison to nmr data. Journal of Computational Chemistry , 34(25):2135–2145, 2013

    work page 2013

  68. [68]

    Charmm36m: an improved force field for folded and intrinsically disordered proteins

    Jing Huang, Sarah Rauscher, Grzegorz Nawrocki, Ting Ra n, Michael Feig, Bert L De Groot, Helmut Grubm¨ uller, and Alexander D MacKerell Jr. Charmm36m: an improved force field for folded and intrinsically disordered proteins. Nature Methods, 14(1):71–73, 2017

    work page 2017

  69. [69]

    Charmm-gui: a web-based graphical user interface for charmm

    Sunhwan Jo, Taehoon Kim, Vidyashankara G Iyer, and Wonp il Im. Charmm-gui: a web-based graphical user interface for charmm. Journal of Computational Chemistry , 29(11):1859–1865, 2008

    work page 2008

  70. [70]

    Charmm-gui input generator for namd, gromacs, amber, openmm, and charm m/openmm simulations using the charmm36 additive force field

    Jumin Lee, Xi Cheng, Sunhwan Jo, Alexander D MacKerell, Jeffery B Klauda, and Wonpil Im. Charmm-gui input generator for namd, gromacs, amber, openmm, and charm m/openmm simulations using the charmm36 additive force field. Biophysical journal , 110(3):641a, 2016

    work page 2016

  71. [71]

    Dimensions of biological cellulose nanocrystals maximize fracture strength

    Robert Sinko, Shawn Mishra, Luis Ruiz, Nick Brandis, an d Sinan Keten. Dimensions of biological cellulose nanocrystals maximize fracture strength. ACS Macro Letters , 3(1):64–69, 2014. 25

    work page 2014

  72. [72]

    Single-molecule investigations of sing le-chain cellulose biosynthesis

    Mark A Hilton, Harris W Manning, Ireneusz G´ orniak, Son ia K Brady, Madeline M Johnson, Jochen Zimmer, and Matthew J Lang. Single-molecule investigations of sing le-chain cellulose biosynthesis. Proceedings of the National Academy of Sciences , 119(40):e2122770119, 2022

    work page 2022

  73. [73]

    Estimation of the persistence length of polymers by md simulations on smal l fragments in solution

    Loes MJ Kroon-Batenburg, Peter H Kruiskamp, Johannes F G Vliegenthart, and Jan Kroon. Estimation of the persistence length of polymers by md simulations on smal l fragments in solution. application to cellulose. The Journal of Physical Chemistry B , 101(42):8454–8459, 1997

    work page 1997

  74. [74]

    Nanoscale structural and me- chanical characterization of a natural nanocomposite mate rial: the shell of red abalone

    Xiaodong Li, Wei-Che Chang, Yuh J Chao, Rizhi Wang, and M ing Chang. Nanoscale structural and me- chanical characterization of a natural nanocomposite mate rial: the shell of red abalone. Nano Letters , 4(4):613–617, 2004

    work page 2004

  75. [75]

    Bioinspired bou ligand cellulose nanocrystal composites: a review of mechanical properties

    Bharath Natarajan and Jeffrey W Gilman. Bioinspired bou ligand cellulose nanocrystal composites: a review of mechanical properties. Philosophical Transactions of the Royal Society A: Mathema tical, Physical and Engineering Sciences, 376(2112):20170050, 2018

    work page 2018

  76. [76]

    Reach coarse-grained simulation of a cellulose fiber

    Dennis C Glass, Kei Moritsugu, Xiaolin Cheng, and Jerem y C Smith. Reach coarse-grained simulation of a cellulose fiber. Biomacromolecules, 13(9):2634–2644, 2012

    work page 2012

  77. [77]

    Hydration control of the mechanical and dynamical properties of cellulose

    Loukas Petridis, Hugh M O’Neill, Mariah Johnsen, Bingx in Fan, Roland Schulz, Eugene Mamontov, Janna Maranas, Paul Langan, and Jeremy C Smith. Hydration control of the mechanical and dynamical properties of cellulose. Biomacromolecules, 15(11):4152–4159, 2014

    work page 2014

  78. [78]

    Strengthening and tou ghening hierarchical nanocellulose via humidity-mediated interface

    YuanZhen Hou, Qing-Fang Guan, Jun Xia, Zhang-Chi Ling, ZeZhou He, Zi-Meng Han, Huai-Bin Yang, Ping Gu, YinBo Zhu, Shu-Hong Yu, et al. Strengthening and tou ghening hierarchical nanocellulose via humidity-mediated interface. ACS nano , 15(1):1310–1320, 2020

    work page 2020

  79. [79]

    Molecular dyn amics study on mechanical properties of cellulose with water molecules diffusion behavior at different oxygen c oncentrations

    Yuanyuan Guo, Wei Wang, and Xuewei Jiang. Molecular dyn amics study on mechanical properties of cellulose with water molecules diffusion behavior at different oxygen c oncentrations. Forests, 14(2):371, 2023. Supplementary Information The following subsections and data are included in the Supplement ary Information. Bonded properties without rescale Force a...

    work page 2023