pith. sign in

arxiv: 2606.27636 · v1 · pith:3LFH4PNCnew · submitted 2026-06-26 · ❄️ cond-mat.mtrl-sci

CoTAR: Topology and Atomic State Reconstruction in Condensed Phases

Hodaka Mori , Yu Miyazaki , Takechika Kikkawa This is my paper

Pith reviewed 2026-06-29 00:41 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci
keywords topology reconstructiongraph neural networkhidden Markov modelmachine learning interatomic potentialscondensed phasemolecular dynamicsbond orderatomic state
0
0 comments X

The pith

CoTAR reconstructs molecular topologies, formal charges, and unpaired electrons from atomic species, coordinates, and total charge using a hybrid GNN-HMM framework.

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

The paper develops CoTAR to supply explicit molecular topology to condensed-phase simulations run with universal machine learning interatomic potentials. These potentials deliver accurate dynamics yet omit bond information required for bond-aware analysis or reconnection to classical force fields. CoTAR performs message passing over a proximity graph, incorporates a van der Waals prior and chemical constraints, and applies hidden Markov model smoothing across time. On classical MD trajectories from 128 nonreactive systems it reaches a bond-order-weighted F1 of 0.906; few-shot fine-tuning lifts the valid-snapshot rate on uMLIP data from 38.6 percent to 84.7 percent, and the resulting topologies support further classical MD runs.

Core claim

CoTAR is a hybrid graph neural network and hidden Markov model that reconstructs molecular topology, formal charges, and unpaired electrons by message passing on a proximity graph augmented by a van der Waals prior and chemical constraints, followed by temporal smoothing; the framework yields a bond-order-weighted F1 score of 0.906 across 128 nonreactive condensed-phase systems on classical MD data and raises the fraction of valid uMLIP snapshots from 38.6 percent to 84.7 percent after few-shot fine-tuning.

What carries the argument

The CoTAR hybrid GNN-HMM framework that performs message passing on proximity graphs together with a van der Waals prior, chemical constraints, and temporal smoothing.

If this is right

  • Reconstructed topologies enable bond-aware analysis of uMLIP trajectories.
  • Few-shot fine-tuning raises the valid-snapshot rate on uMLIP data from 38.6 percent to 84.7 percent.
  • The topologies support downstream classical MD simulations.
  • HMM smoothing increases system-level MD simulation feasibility from 83.6 percent to 85.9 percent.

Where Pith is reading between the lines

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

  • The method could link uMLIP dynamics directly to existing classical force-field pipelines without manual topology assignment.
  • Relaxing the nonreactive assumption might allow the same reconstruction machinery to handle bond-breaking events.
  • Analogous proximity-graph plus constraint models could be tested on other particle simulations that lack explicit connectivity.

Load-bearing premise

The 128 tested nonreactive systems represent the condensed-phase cases where uMLIP trajectories are used, and the combination of proximity-graph message passing, van der Waals prior, and chemical constraints produces chemically valid topologies without further system-specific tuning.

What would settle it

Applying CoTAR to uMLIP trajectories from a condensed-phase system outside the original 128 and observing that few-shot fine-tuning leaves the valid-snapshot rate near 38.6 percent would show the reconstruction does not generalize.

Figures

Figures reproduced from arXiv: 2606.27636 by Hodaka Mori, Takechika Kikkawa, Yu Miyazaki.

Figure 1
Figure 1. Figure 1: Overview of CoTAR. A proximity [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Downstream classical-MD validation using topologies predicted by the full CoTAR pipeline. The horizontal axis is the density ob￾tained from the reference topology, ρoriginal, and the vertical axis is the density obtained from the predicted topology, ρreconstruct. Points are colored by category, error bars denote ±1σ, and the legends report category-specific MD￾feasibility rates. rather than the primary sou… view at source ↗
Figure 3
Figure 3. Figure 3: Inference time per snapshot versus atom count for a separate liquid-toluene scaling test. The atom count was varied by changing the number of toluene molecules in the periodic liquid cell. The same configurations were eval￾uated with all methods shown in the legend. This figure is intended only to illustrate com￾putational scaling and is not part of the 128- system accuracy benchmark. cost [PITH_FULL_IMAG… view at source ↗
read the original abstract

Universal machine learning interatomic potentials (uMLIPs) enable condensed-phase molecular dynamics (MD) simulations with near-first-principles accuracy, but their lack of explicit molecular topology limits bond-aware analysis and reconnection to classical force fields. Here, we present CoTAR, a hybrid graph neural network (GNN)--hidden Markov model (HMM) framework that reconstructs molecular topology, formal charges, and unpaired electrons from atomic species, coordinates, and total charge by combining message passing on a proximity graph with a van der Waals prior, chemical constraints, and temporal smoothing. Across 128 nonreactive, topology-preserving condensed-phase systems, CoTAR achieved a bond-order-weighted F1 score of 0.906 on classical-MD data; for uMLIP trajectories, few-shot fine-tuning improved the valid-snapshot rate from 38.6\% to 84.7\%. The reconstructed topologies also supported downstream classical MD simulations, and HMM smoothing improved system-level MD simulation feasibility from 83.6\% to 85.9\%, indicating that CoTAR provides a practical framework for bond-aware analysis of condensed-phase uMLIP trajectories.

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

0 major / 3 minor

Summary. The manuscript introduces CoTAR, a hybrid GNN-HMM framework that reconstructs molecular topology, formal charges, and unpaired electrons from atomic species, coordinates, and total charge by combining message passing on a proximity graph with a van der Waals prior, chemical constraints, and temporal smoothing. Across 128 nonreactive, topology-preserving condensed-phase systems, it reports a bond-order-weighted F1 score of 0.906 on classical-MD data; few-shot fine-tuning on uMLIP trajectories improves the valid-snapshot rate from 38.6% to 84.7%. The reconstructed topologies support downstream classical MD simulations, and HMM smoothing raises system-level MD feasibility from 83.6% to 85.9%.

Significance. If the reported metrics are robust, CoTAR would address a practical gap in uMLIP usage by enabling bond-aware analysis and reconnection to classical force fields without system-specific tuning. The scale of testing (128 systems) and the quantified improvement in valid snapshots constitute a concrete contribution to the field.

minor comments (3)
  1. [Abstract] Abstract: the dataset composition, selection criteria, and diversity metrics for the 128 systems are not described, which would help readers assess representativeness of the tested condensed-phase cases.
  2. [Abstract] Abstract: performance numbers are given without error bars, standard deviations, or details on train/test splits and ablation studies; adding these would strengthen the presentation of the F1 and valid-snapshot results.
  3. The manuscript would benefit from a brief comparison table or section contrasting CoTAR against existing topology-reconstruction methods (e.g., rule-based or other GNN approaches) to clarify the incremental advance.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of CoTAR, the recognition of its practical utility for uMLIP trajectories, and the recommendation for minor revision. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The abstract and available description present CoTAR as a hybrid GNN-HMM method whose performance metrics (bond-order-weighted F1 of 0.906 on 128 systems; valid-snapshot rate improvement from 38.6% to 84.7% after few-shot fine-tuning) are reported as direct empirical measurements on classical-MD and uMLIP trajectories. No equations, parameter-fitting steps, or self-citations appear in the supplied text that would reduce any claimed prediction or reconstruction result to a tautology by construction. The method description (proximity-graph message passing plus van der Waals prior plus chemical constraints plus temporal smoothing) is stated at a level that does not exhibit self-definitional, fitted-input, or self-citation-load-bearing circularity. The evaluation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The abstract supplies no explicit free parameters, axioms, or invented entities beyond the standard assumptions of graph neural networks and hidden Markov models. The van der Waals prior and chemical constraints are invoked but not quantified.

axioms (1)
  • domain assumption Message passing on a proximity graph combined with HMM temporal smoothing can recover chemically valid topologies when supplemented by a van der Waals prior and chemical constraints.
    This modeling choice is central to the CoTAR framework described in the abstract.

pith-pipeline@v0.9.1-grok · 5732 in / 1434 out tokens · 36487 ms · 2026-06-29T00:41:43.502891+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

97 extracted references · 4 canonical work pages

  1. [1]

    Accelerated ReaxFF Simulations for Describing the Reactive Cross-Linking of Polymers

    Vashisth, Aniruddh and Ashraf, Chowdhury and Zhang, Weiwei and Bakis, Charles E and van Duin, Adri C T. Accelerated ReaxFF Simulations for Describing the Reactive Cross-Linking of Polymers. J. Phys. Chem. A

  2. [2]

    Simulations of the biodegradation of citrate-based polymers for artificial scaffolds using accelerated reactive molecular dynamics

    Dasgupta, Nabankur and Yilmaz, Dundar E and van Duin, Adri. Simulations of the biodegradation of citrate-based polymers for artificial scaffolds using accelerated reactive molecular dynamics. J. Phys. Chem. B

  3. [3]

    ReaxFF : A Reactive Force Field for Hydrocarbons

    van Duin, Adri C T and Dasgupta, Siddharth and Lorant, Francois and Goddard, William A. ReaxFF : A Reactive Force Field for Hydrocarbons. J. Phys. Chem. A

  4. [4]

    Escaping free-energy minima

    Laio, Alessandro and Parrinello, Michele. Escaping free-energy minima. Proc. Natl. Acad. Sci. U. S. A

  5. [5]

    Ilyes Batatia and David Peter Kovacs and Gregor N. C. Simm and Christoph Ortner and Gabor Csanyi , booktitle=. 2022 , url=

    2022

  6. [6]

    Batatia, S

    The Design Space of E(3)-Equivariant Atom-Centered Interatomic Potentials , author =. 2022 , number =. doi:10.48550/arXiv.2205.06643 , archiveprefix =. 2205.06643 , eprinttype =

    work page doi:10.48550/arxiv.2205.06643 2022

  7. [7]

    CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling

    Deng, Bowen and Zhong, Peichen and Jun, Kyujung and Riebesell, Janosh and Han, Kevin and Bartel, Christopher J and Ceder, Gerbrand. CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling. Nature Machine Intelligence

  8. [8]

    Towards universal neural network potential for material discovery applicable to arbitrary combination of 45 elements

    Takamoto, So and Shinagawa, Chikashi and Motoki, Daisuke and Nakago, Kosuke and Li, Wenwen and Kurata, Iori and Watanabe, Taku and Yayama, Yoshihiro and Iriguchi, Hiroki and Asano, Yusuke and Onodera, Tasuku and Ishii, Takafumi and Kudo, Takao and Ono, Hideki and Sawada, Ryohto and Ishitani, Ryuichiro and Ong, Marc and Yamaguchi, Taiki and Kataoka, Toshik...

  9. [9]

    A universal graph deep learning interatomic potential for the periodic table

    Chen, Chi and Ong, Shyue Ping. A universal graph deep learning interatomic potential for the periodic table. Nat. Comput. Sci

  10. [10]

    The potential of neural network potentials

    Duignan, Timothy T. The potential of neural network potentials. ACS Phys. Chem. Au

  11. [11]

    Neural network potentials for chemistry: concepts, applications and prospects

    Käser, Silvan and Vazquez-Salazar, Luis Itza and Meuwly, Markus and Töpfer, Kai. Neural network potentials for chemistry: concepts, applications and prospects. Digit. Discov

  12. [12]

    An overview about neural networks potentials in molecular dynamics simulation

    Martin-Barrios, Raidel and Navas-Conyedo, Edisel and Zhang, Xuyi and Chen, Yunwei and Gulín-González, Jorge. An overview about neural networks potentials in molecular dynamics simulation. Int. J. Quantum Chem

  13. [13]

    Simulations of glass transition and mechanical behavior of off-stoichiometric crosslinked polymers

    Bezik, Cody T and Redline, Erica M and Foster, Jeffrey C and Frischknecht, Amalie L. Simulations of glass transition and mechanical behavior of off-stoichiometric crosslinked polymers. Macromolecules

  14. [14]

    Verification for Temperature Dependence of Tacticity in Polystyrene Radical Polymerization with the Combination of Reaction Pathway Analysis and Red Moon Methodology

    Rao, Zizhen and Takayanagi, Masayoshi and Nagaoka, Masataka. Verification for Temperature Dependence of Tacticity in Polystyrene Radical Polymerization with the Combination of Reaction Pathway Analysis and Red Moon Methodology. J. Phys. Chem. B

  15. [15]

    Molecular modeling of reactive systems with REACTER

    Gissinger, Jacob R and Jensen, Benjamin D and Wise, Kristopher E. Molecular modeling of reactive systems with REACTER. Comput. Phys. Commun

  16. [16]

    Enhancing epoxy resin curing: Investigating the catalytic role of water as a trace impurity in dense crosslinked network formation using an advanced cat- GRRM / MC / MD Method1

    Xi, Yingxiao and Fukuzawa, Hironobu and Kikugawa, Gota and Zhao, Yinbo and Kawagoe, Yoshiaki and Okabe, Tomonaga and Kishi, Hajime and Kishimoto, Naoki. Enhancing epoxy resin curing: Investigating the catalytic role of water as a trace impurity in dense crosslinked network formation using an advanced cat- GRRM / MC / MD Method1. Polymer (Guildf.)

  17. [17]

    Development of cat- GRRM / MC / MD method for the simulation of cross-linked network structure formation with molecular autocatalysis

    Xi, Yingxiao and Fukuzawa, Hironobu and Fukunaga, Shoji and Kikugawa, Gota and Zhao, Yinbo and Kawagoe, Yoshiaki and Okabe, Tomonaga and Kishimoto, Naoki. Development of cat- GRRM / MC / MD method for the simulation of cross-linked network structure formation with molecular autocatalysis. Molecular Catalysis

  18. [18]

    Curing reaction of epoxy resin composed of mixed base resin and curing agent: Experiments and molecular simulation

    Okabe, Tomonaga and Takehara, Tomohiro and Inose, Keisuke and Hirano, Noriyuki and Nishikawa, Masaaki and Uehara, Takuya. Curing reaction of epoxy resin composed of mixed base resin and curing agent: Experiments and molecular simulation. Polymer

  19. [19]

    Chemical reactions in classical molecular dynamics

    Gissinger, Jacob R and Jensen, Benjamin D and Wise, Kristopher E. Chemical reactions in classical molecular dynamics. Polymer

  20. [20]

    Kinetics of the interfacial curing reaction for an epoxy-amine mixture

    Yamaguchi, Ko and Kawaguchi, Daisuke and Miyata, Noboru and Miyazaki, Tsukasa and Aoki, Hiroyuki and Yamamoto, Satoru and Tanaka, Keiji. Kinetics of the interfacial curing reaction for an epoxy-amine mixture. Phys. Chem. Chem. Phys

  21. [21]

    Effect of water content on the thermal degradation of amorphous polyamide 6,6: A collective variable-driven hyperdynamics study

    Arash, Behrouz and Thijsse, Barend J and Pecenko, Alessandro and Simone, Angelo. Effect of water content on the thermal degradation of amorphous polyamide 6,6: A collective variable-driven hyperdynamics study. Polym. Degrad. Stab

  22. [22]

    Resin filling into nano-sized pore on metal surface analyzed by all-atom molecular dynamics simulation over a variety of resin and pore sizes

    Mori, Hodaka and Matubayasi, Nobuyuki. Resin filling into nano-sized pore on metal surface analyzed by all-atom molecular dynamics simulation over a variety of resin and pore sizes. Polymer

  23. [23]

    Molecular dynamics modeling of epoxy resins using the reactive interface force field

    Odegard, Gregory M and Patil, Sagar U and Deshpande, Prathamesh P and Kanhaiya, Krishan and Winetrout, Jordan J and Heinz, Hendrik and Shah, Sagar P and Maiaru, Marianna. Molecular dynamics modeling of epoxy resins using the reactive interface force field. Macromolecules

  24. [24]

    A review on molecularly imprinted polymers preparation by computational simulation-aided methods

    Liu, Zhimin and Xu, Zhigang and Wang, Dan and Yang, Yuming and Duan, Y and Ma, Liping and Lin, Tao and Liu, Hongcheng. A review on molecularly imprinted polymers preparation by computational simulation-aided methods. Polymers

  25. [25]

    Molecular dynamics simulation of polyamide-based materials – A review

    Krishna, Sanjay and Sreedhar, I and Patel, Chetan M. Molecular dynamics simulation of polyamide-based materials – A review. Comput. Mater. Sci

  26. [26]

    A review of advancements in coarse-grained molecular dynamics simulations

    Joshi, Soumil Y and Deshmukh, Sanket A. A review of advancements in coarse-grained molecular dynamics simulations. Mol. Simul

  27. [27]

    Modeling and simulations of polymers: A roadmap

    Gartner, III, Thomas E and Jayaraman, Arthi. Modeling and simulations of polymers: A roadmap. Macromolecules

  28. [28]

    Rate coefficients of free-radical polymerization deduced from pulsed laser experiments

    Beuermann, Sabine and Buback, Michael. Rate coefficients of free-radical polymerization deduced from pulsed laser experiments. Prog. Polym. Sci

  29. [29]

    Cure kinetics of several epoxy–amine systems at ambient and high temperatures

    Pramanik, Monoj and Fowler, Eric W and Rawlins, James W. Cure kinetics of several epoxy–amine systems at ambient and high temperatures. J Coat Technol Res

  30. [30]

    Theoretical catalyst screening of multielement alloy catalysts for ammonia synthesis using machine learning potential and generative artificial intelligence

    Hisama, Kaoru and Ishikawa, Atsushi and Aspera, Susan Menez and Koyama, Michihisa. Theoretical catalyst screening of multielement alloy catalysts for ammonia synthesis using machine learning potential and generative artificial intelligence. J. Phys. Chem. C Nanomater. Interfaces

  31. [31]

    Interconnected lamellar 3D semiconductive PCP for rechargeable aqueous zinc battery cathodes

    Lin, Zirui and Otake, Ken-Ichi and Kajiwara, Takashi and Hiraide, Shotaro and Nurhuda, Maryam and Packwood, Daniel and Kadota, Kentaro and Sakamoto, Hirotoshi and Kawaguchi, Shogo and Kubota, Yoshiki and Yao, Ming-Shui and Horike, Satoshi and Sun, Xiaoqi and Kitagawa, Susumu. Interconnected lamellar 3D semiconductive PCP for rechargeable aqueous zinc batt...

  32. [32]

    Molecular dynamics of liquid-electrode interface by integrating Coulomb interaction into universal neural network potential

    Hisama, Kaoru and Valadez Huerta, Gerardo and Koyama, Michihisa. Molecular dynamics of liquid-electrode interface by integrating Coulomb interaction into universal neural network potential. J. Comput. Chem

  33. [33]

    A neural network potential for the IRMOF series and its application for thermal and mechanical behaviors

    Tayfuroglu, Omer and Kocak, Abdulkadir and Zorlu, Y. A neural network potential for the IRMOF series and its application for thermal and mechanical behaviors. Phys. Chem. Chem. Phys

  34. [34]

    Polymer Science and Technology

    Ebewele, Robert O. Polymer Science and Technology

  35. [35]

    Introduction to Polymers

    Young, Robert J and Lovell, Peter A. Introduction to Polymers

  36. [36]

    Handbook of polymers

    Wypych, G. Handbook of polymers

  37. [37]

    Principles of polymer engineering

    McCrum, N G and Buckley, C P and Bucknall, C B. Principles of polymer engineering

  38. [38]

    Rethinking metadynamics: From bias potentials to probability distributions

    Invernizzi, Michele and Parrinello, Michele. Rethinking metadynamics: From bias potentials to probability distributions. J. Phys. Chem. Lett

  39. [39]

    1999 , publisher=

    Polymer handbook , author=. 1999 , publisher=

    1999

  40. [40]

    Merging metadynamics into hyperdynamics: accelerated molecular simulations reaching time scales from microseconds to seconds

    Bal, Kristof M and Neyts, Erik C. Merging metadynamics into hyperdynamics: accelerated molecular simulations reaching time scales from microseconds to seconds. J. Chem. Theory Comput

  41. [41]

    Hyperdynamics: Accelerated molecular dynamics of infrequent events

    Voter, Arthur F. Hyperdynamics: Accelerated molecular dynamics of infrequent events. Phys. Rev. Lett

  42. [42]

    Accelerated molecular dynamics with the bond-boost method

    Miron, Radu A and Fichthorn, Kristen A. Accelerated molecular dynamics with the bond-boost method. J. Chem. Phys

  43. [43]

    Stukowski, Alexander , Title =

  44. [44]

    VMD : visual molecular dynamics

    Humphrey, W and Dalke, A and Schulten, K. VMD : visual molecular dynamics. J. Mol. Graph

  45. [45]

    2013 , publisher=

    Fundamentals of controlled/living radical polymerization , author=. 2013 , publisher=

    2013

  46. [46]

    2004 , publisher=

    Principles of polymerization , author=. 2004 , publisher=

    2004

  47. [47]

    Chemical Science , volume=

    Intramolecular trapping of spiro radicals to produce unusual cyclization products from usual migration substrates , author=. Chemical Science , volume=. 2023 , publisher=

    2023

  48. [48]

    European polymer journal , volume=

    Effect of 1, 1-diphenylethylene on the radical polymerization of di-n-butyl itaconate in benzene , author=. European polymer journal , volume=. 2001 , publisher=

    2001

  49. [49]

    2016 , publisher=

    CRC handbook of chemistry and physics , author=. 2016 , publisher=

    2016

  50. [50]

    The Journal of Physical Chemistry A , volume=

    Accelerated ReaxFF simulations for describing the reactive cross-linking of polymers , author=. The Journal of Physical Chemistry A , volume=. 2018 , publisher=

    2018

  51. [51]

    The Journal of Physical Chemistry B , volume=

    OpenMM 8: molecular dynamics simulation with machine learning potentials , author=. The Journal of Physical Chemistry B , volume=. 2023 , publisher=

    2023

  52. [52]

    The Journal of Physical Chemistry C , volume=

    DFT simulation of XPS reveals Cu/epoxy polymer interfacial bonding , author=. The Journal of Physical Chemistry C , volume=. 2019 , publisher=

    2019

  53. [53]

    ACS Applied Polymer Materials , volume=

    Molecular events for an epoxy--amine system at a copper interface , author=. ACS Applied Polymer Materials , volume=. 2020 , publisher=

    2020

  54. [54]

    Chemical Physics Letters , volume=

    A hybrid MC/MD reaction method with rare event-driving mechanism: Atomistic realization of 2-chlorobutane racemization process in DMF solution , author=. Chemical Physics Letters , volume=. 2013 , publisher=

    2013

  55. [55]

    The Journal of Chemical Physics , volume=

    A transformation theory of stochastic evolution in Red Moon methodology to time evolution of chemical reaction process in the full atomistic system , author=. The Journal of Chemical Physics , volume=. 2017 , publisher=

    2017

  56. [56]

    1,1-Diphenylethylene (97\ url =

  57. [57]

    Matlantis, software as a service style material discovery tool

  58. [58]

    The Journal of Physical Chemistry C , volume=

    Development of a reactive force field for simulations on the catalytic conversion of C/H/O molecules on Cu-metal and Cu-oxide surfaces and application to Cu/CuO-based chemical looping , author=. The Journal of Physical Chemistry C , volume=. 2020 , publisher=

    2020

  59. [59]

    The Journal of Physical Chemistry A , volume=

    Computational Insights into Tunable Reversible Network Materials: Accelerated ReaxFF Kinetics of Furan-Maleimide Diels--Alder Reactions for Self-Healing and Recyclability , author=. The Journal of Physical Chemistry A , volume=. 2024 , publisher=

    2024

  60. [60]

    Journal of molecular graphics , volume=

    VMD: visual molecular dynamics , author=. Journal of molecular graphics , volume=. 1996 , publisher=

    1996

  61. [61]

    Modelling and simulation in materials science and engineering , volume=

    Visualization and analysis of atomistic simulation data with OVITO--the Open Visualization Tool , author=. Modelling and simulation in materials science and engineering , volume=

  62. [62]

    Bulletin of the Korean Chemical Society , volume=

    Universal structure conversion method for organic molecules: from atomic connectivity to three-dimensional geometry , author=. Bulletin of the Korean Chemical Society , volume=. 2015 , publisher=

    2015

  63. [63]

    Journal of Chemical Information and Modeling , volume=

    Multimodal Bond Reconstruction toward Generative Molecular Design , author=. Journal of Chemical Information and Modeling , volume=. 2026 , publisher=

    2026

  64. [66]

    Physical chemistry chemical physics , volume=

    ReacNetGenerator: an automatic reaction network generator for reactive molecular dynamics simulations , author=. Physical chemistry chemical physics , volume=. 2020 , publisher=

    2020

  65. [67]

    Machine learning meets quantum physics , pages=

    Message passing neural networks , author=. Machine learning meets quantum physics , pages=. 2020 , publisher=

    2020

  66. [68]

    Current opinion in structural biology , volume=

    Hidden markov models , author=. Current opinion in structural biology , volume=. 1996 , publisher=

    1996

  67. [69]

    IEEE transactions on communications , volume=

    List Viterbi decoding algorithms with applications , author=. IEEE transactions on communications , volume=. 1994 , publisher=

    1994

  68. [70]

    mendeleev: A Python resource for properties of chemical elements, ions and isotopes , year =

    Mentel,. mendeleev: A Python resource for properties of chemical elements, ions and isotopes , year =

  69. [71]

    RDKit: Open-Source Cheminformatics Software , url =

    Landrum, Greg , biburl =. RDKit: Open-Source Cheminformatics Software , url =

  70. [72]

    0: the sage small molecule force field , author=

    Development and benchmarking of open force field 2.0. 0: the sage small molecule force field , author=. Journal of chemical theory and computation , volume=. 2023 , publisher=

    2023

  71. [73]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Towards understanding convergence and generalization of AdamW , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2024 , publisher=

    2024

  72. [74]

    Journal of Cheminformatics , volume=

    A rule-based algorithm for automatic bond type perception , author=. Journal of Cheminformatics , volume=. 2012 , publisher=

    2012

  73. [75]

    ChemRxiv , volume =

    Christoph Loschen , title =. ChemRxiv , volume =. 2018 , doi =. https://chemrxiv.org/doi/pdf/10.26434/chemrxiv.7403630.v2 , abstract =

    work page doi:10.26434/chemrxiv.7403630.v2 2018

  74. [76]

    Journal of Applied Physics , volume=

    Bond order predictions using deep neural networks , author=. Journal of Applied Physics , volume=. 2021 , publisher=

    2021

  75. [77]

    P.; Simm, G

    Batatia, I.; Kovacs, D. P.; Simm, G. N. C.; Ortner, C.; Csanyi, G. MACE : Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields. Advances in Neural Information Processing Systems. 2022

    2022

  76. [78]

    P.; Musaelian, A.; Simm, G

    Batatia, I.; Batzner, S.; Kov \'a cs, D. P.; Musaelian, A.; Simm, G. N. C.; Drautz, R.; Ortner, C.; Kozinsky, B.; Cs \'a nyi, G. The Design Space of E(3)-Equivariant Atom-Centered Interatomic Potentials. 2022

    2022

  77. [79]

    Takamoto, S. et al. Towards universal neural network potential for material discovery applicable to arbitrary combination of 45 elements. Nat. Commun. 2022, 13, 2991

    2022

  78. [80]

    Chen, C.; Ong, S. P. A universal graph deep learning interatomic potential for the periodic table. Nat. Comput. Sci. 2022, 2, 718--728

    2022

  79. [81]

    J.; Ceder, G

    Deng, B.; Zhong, P.; Jun, K.; Riebesell, J.; Han, K.; Bartel, C. J.; Ceder, G. CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling. Nature Machine Intelligence 2023, 5, 1031--1041

    2023

  80. [82]

    M.; Dzamba, M.; Fu, X.; Gao, M.; Shuaibi, M.; Barroso-Luque, L.; Abdelmaqsoud, K.; Gharakhanyan, V.; Kitchin, J

    Wood, B. M.; Dzamba, M.; Fu, X.; Gao, M.; Shuaibi, M.; Barroso-Luque, L.; Abdelmaqsoud, K.; Gharakhanyan, V.; Kitchin, J. R.; Levine, D. S.; others Uma: A family of universal models for atoms. arXiv preprint arXiv:2506.23971 2025,

    work page arXiv 2025

Showing first 80 references.