arxiv: 2604.21494 · v1 · submitted 2026-04-23 · ❄️ cond-mat.mtrl-sci · physics.chem-ph

Recognition: unknown

Data-Driven Thermal and Mechanical Modeling of Defective Covalent Organic Frameworks

Aleksander Szewczyk , Leonardo Medrano Sandonas , David Bodesheim , Bohayra Mortazavi , Gianaurelio Cuniberti

Authors on Pith no claims yet

Pith reviewed 2026-05-09 20:50 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci physics.chem-ph

keywords covalent organic frameworksmachine learning interatomic potentialsdefective COFsthermal conductivitymolecular dynamicsdensity functional theoryphonon dispersionmechanical properties

0 comments

The pith

An invariant QCOF model with small descriptor size outperforms general machine learning potentials for simulating defective covalent organic frameworks.

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

The authors train specialized QCOF machine learning interatomic potentials by tuning the MACE architecture on extensive non-equilibrium density functional theory data for covalent organic frameworks. An invariant version with reduced descriptor dimensionality and cutoff radius outperforms general-purpose MACE models and fine-tuned variants across benchmarks including force accuracy in defective structures, system scalability, and phonon dispersion. This model then powers large-scale non-equilibrium molecular dynamics runs that quantify thermal conductivity drops in defective CTF-1 and COF-LZU1, finding stronger defect sensitivity in CTF-1, alongside stress-strain curves that remain nearly unchanged at low defect levels but turn asymmetric at large strains.

Core claim

The paper establishes that an invariant QCOF model with small descriptor dimensionality and cutoff outperforms all other models in most validation tasks, including scalability to large systems, force prediction in defective COFs, and phonon dispersion calculations. The best-performing QCOF model was then used to run large-scale simulations of thermal conductivity for defective CTF-1 and COF-LZU1 systems via non-equilibrium MD, revealing a more pronounced sensitivity of CTF-1 to structural defects, while stress-strain curves show nearly invariant mechanical response at low defect densities with asymmetric behaviour at large strains.

What carries the argument

The invariant QCOF model, a tuned MACE architecture with small descriptor dimensionality and cutoff trained on non-equilibrium DFT COF conformations.

Load-bearing premise

Machine learning interatomic potentials trained on non-equilibrium DFT conformations of COFs will accurately capture thermal and mechanical behavior in large defective systems outside the training distribution.

What would settle it

A direct measurement of thermal conductivity in a synthesized defective CTF-1 sample compared against the non-equilibrium MD prediction from the QCOF model.

read the original abstract

Covalent Organic Frameworks (COFs) are versatile two-dimensional (2D) materials for flexible electronics, catalysis, and sensing, owing to their tunable architectures and large surface areas. However, like most materials, COFs inevitably contain synthesis-induced defects, which-similar to graphene-can strongly influence intrinsic properties, such as thermal transport and mechanical strength. To address this challenge, we have assessed the performance of a set of machine learning interatomic potentials (MLIP) capable of efficient large-scale simulations of COFs with quantum accuracy. In doing so, QCOF models (Quantum COF) were developed by tuning the state-of-the-art MACE architecture on an extensive dataset of non-equilibrium COF conformations generated from high-fidelity density functional theory calculations. The accuracy, computational efficiency, memory footprint, and transferability to unseen chemical environments of these models were benchmarked against general-purpose MACE models and their fine-tuned variants. Our results show that an invariant QCOF model with a small descriptor dimensionality and cutoff outperforms all other models in most validation tasks, including scalability to large systems, force prediction in defective COFs, and phonon dispersion calculations. The best-performing QCOF model was then used to run large-scale simulations of thermal conductivity for defective CTF-1 and COF-LZU1 systems via non-equilibrium MD, revealing a more pronounced sensitivity of CTF-1 to structural defects. Stress-strain curves were also investigated, showing that the mechanical response remains nearly invariant at low defect densities, while asymmetric behaviour emerges at large strains. This work thus provides a foundation for the design of robust quantum-informed MLIP for large-scale property simulations of defective of extended network materials.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The paper tunes MACE into COF-specific models that beat the general version on force accuracy and scalability for defective structures, then uses the best one for large NE-MD runs showing CTF-1 thermal conductivity drops more with defects than COF-LZU1.

read the letter

The core advance is the creation of QCOF variants by retraining MACE on DFT-generated non-equilibrium COF conformations, followed by direct benchmarks on defective test sets and then large-scale simulations of thermal transport and stress-strain response. They report that a compact invariant model with reduced descriptor size and cutoff wins on most tasks, including force errors in defective COFs and phonon calculations, and they apply it to CTF-1 and COF-LZU1 supercells with varying defect densities via non-equilibrium molecular dynamics. That gives a concrete, practical route to studying defect effects in these 2D networks without full DFT cost at scale. The mechanical part shows low sensitivity at small defect levels and asymmetry only at high strain, which aligns with expectations for network materials. The thermal results highlight material-specific defect sensitivity, which is the kind of finding that could guide synthesis priorities. The work is grounded in standard DFT reference data and existing MACE architecture, with no obvious circularity in the claims. The soft spot is that the headline thermal-conductivity differences rest on the model's extrapolation to defect-induced scattering without a direct DFT or Boltzmann-transport cross-check on even modestly sized defective cells for the conductivity values themselves. Force accuracy on defects is shown, but anharmonic transport can be sensitive to small errors in new local environments not heavily sampled in training. Phonon dispersions are likely reported on pristine cells. If the full paper includes those extra checks or error bars on the conductivity reductions, the concern shrinks; otherwise the quantitative sensitivity statements stay provisional. This is useful for computational materials groups focused on porous 2D frameworks or ML interatomic potentials for extended networks. A reader already working with MACE or similar equivariant models will pick up the tuning details and the defect-application example quickly. It is solid enough on method and data generation to warrant peer review rather than desk rejection, though referees will probably ask for the missing transport validation step.

Referee Report

3 major / 2 minor

Summary. The manuscript develops a family of MACE-based machine learning interatomic potentials (termed QCOF models) trained on extensive DFT-generated non-equilibrium conformations of covalent organic frameworks. It benchmarks an invariant QCOF variant with reduced descriptor dimensionality and cutoff radius against general-purpose MACE models and fine-tuned variants, reporting superior performance in force accuracy on defective COFs, computational scalability, and phonon dispersion calculations. The best QCOF model is subsequently deployed in large-scale non-equilibrium molecular dynamics to compute thermal conductivity of defective CTF-1 and COF-LZU1, concluding that CTF-1 exhibits stronger defect sensitivity, while stress-strain response remains largely invariant at low defect densities.

Significance. If the central claims hold, the work supplies a practical, quantum-accurate route to simulate thermal transport and mechanics in large defective COF supercells that remain inaccessible to direct DFT. The emphasis on a compact, transferable model and its application to defect-induced changes in conductivity could inform materials design for flexible electronics and catalysis. The use of a large DFT training set and systematic benchmarking against an established architecture are positive features.

major comments (3)

[§5] §5 (thermal conductivity simulations): the headline claim that CTF-1 shows more pronounced defect sensitivity than COF-LZU1 rests on NE-MD trajectories generated with the QCOF model. No direct comparison to ab initio MD or Boltzmann transport equation calculations on even modestly sized defective supercells is provided, leaving the quantitative sensitivity vulnerable to extrapolation error from training data that primarily samples low-defect or pristine environments.
[§4] §4 (benchmarking and validation): while the text asserts that the invariant QCOF model outperforms alternatives in force prediction on defective COFs and phonon calculations, the abstract and main text supply no numerical MAE values, error bars, or tables quantifying the improvement over MACE baselines. This absence prevents assessment of whether the reported outperformance is statistically meaningful or practically significant for the downstream thermal-transport results.
[Methods] Methods (training data generation): the models are trained exclusively on non-equilibrium DFT conformations of (presumably low-defect) COFs. The manuscript does not describe how defect configurations (e.g., missing linkers or vacancies) are sampled or whether any defective structures were included in the training or validation splits, which directly affects the reliability of force predictions and derived transport properties in the defective systems studied later.

minor comments (2)

[Abstract] Abstract, final sentence: the phrasing 'defective of extended network materials' appears to be a typographical error and should be corrected for clarity.
[Figures] Figure captions and tables: several benchmarking figures lack explicit error bars or statistical details on the number of test configurations, which would aid reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. These have helped us identify areas where the manuscript can be clarified and strengthened. We address each major comment point by point below, indicating the revisions we will make in the next version of the manuscript.

read point-by-point responses

Referee: §5 (thermal conductivity simulations): the headline claim that CTF-1 shows more pronounced defect sensitivity than COF-LZU1 rests on NE-MD trajectories generated with the QCOF model. No direct comparison to ab initio MD or Boltzmann transport equation calculations on even modestly sized defective supercells is provided, leaving the quantitative sensitivity vulnerable to extrapolation error from training data that primarily samples low-defect or pristine environments.

Authors: We agree that direct ab initio validation on large defective supercells would be ideal but is computationally infeasible, which is the primary motivation for developing the QCOF MLIPs. Our model was benchmarked on force predictions for defective COF structures (unseen during training) and phonon dispersions, providing indirect support for the NE-MD results. In the revised manuscript we will add a dedicated limitations paragraph discussing extrapolation risks, emphasize the relative (rather than absolute) nature of the CTF-1 vs. COF-LZU1 comparison, and include additional small-system DFT validations of defect-induced force errors in the supplementary information. revision: partial
Referee: §4 (benchmarking and validation): while the text asserts that the invariant QCOF model outperforms alternatives in force prediction on defective COFs and phonon calculations, the abstract and main text supply no numerical MAE values, error bars, or tables quantifying the improvement over MACE baselines. This absence prevents assessment of whether the reported outperformance is statistically meaningful or practically significant for the downstream thermal-transport results.

Authors: We apologize for the lack of prominent numerical values. Detailed MAE tables with error bars and statistical comparisons to MACE baselines are already present in the results section and supplementary material. In the revised version we will extract and highlight the key numerical improvements (force MAE on defective COFs, phonon frequency errors) directly in the abstract and main text so that readers can immediately assess the magnitude and significance of the gains. revision: yes
Referee: Methods (training data generation): the models are trained exclusively on non-equilibrium DFT conformations of (presumably low-defect) COFs. The manuscript does not describe how defect configurations (e.g., missing linkers or vacancies) are sampled or whether any defective structures were included in the training or validation splits, which directly affects the reliability of force predictions and derived transport properties in the defective systems studied later.

Authors: The referee correctly notes that the training set contains only non-equilibrium conformations of pristine COFs; no defective structures were included in training or validation. Defect configurations for the thermal and mechanical studies were generated independently by removing linkers or creating vacancies in large supercells. We will revise the Methods section to state this explicitly, describe the defect-generation protocol, and clarify that the benchmarks on defective COFs test transferability to unseen environments. This information will also be summarized in a new paragraph on model applicability. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard MLIP training-validation-application pipeline

full rationale

The paper trains QCOF models on independent DFT-generated non-equilibrium COF conformations, benchmarks force accuracy, phonon dispersions, and scalability on held-out test sets (including defective structures), and applies the best model to large-scale NE-MD for thermal conductivity and stress-strain curves. No self-definitional loops exist where outputs are defined in terms of themselves; no fitted parameters are relabeled as predictions; validation metrics derive from data partitions external to training; and any self-citations are not load-bearing for the central claims, which rest on new DFT references and comparisons to general MACE. The chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The work depends on DFT as ground-truth reference data and introduces domain-specific ML models without new physical postulates.

free parameters (2)

descriptor dimensionality
Selected as small for the invariant QCOF model to achieve reported outperformance in validation tasks.
cutoff
Chosen as small to optimize the model for COF simulations and scalability.

axioms (1)

domain assumption High-fidelity DFT calculations provide accurate reference data for non-equilibrium COF conformations
Used to generate the extensive training dataset for the QCOF models.

invented entities (1)

QCOF models no independent evidence
purpose: Specialized ML interatomic potentials tuned for COF systems
Fine-tuned variants of the MACE architecture on COF-specific data.

pith-pipeline@v0.9.0 · 5632 in / 1335 out tokens · 44281 ms · 2026-05-09T20:50:27.267547+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

69 extracted references · 64 canonical work pages

[1]

& Yaghi, O

Furukawa, H. & Yaghi, O. M. Storage of Hydrogen, Methane, and Carbon Dioxide in Highly Porous Covalent Organic Frameworks for Clean Energy Applications.J. Am. Chem. Soc.131, 8875–8883, 10.1021/ja9015765 (2009)

work page doi:10.1021/ja9015765 2009
[2]

& Zhang, Q

Wu, J., Zhang, S., Gu, Q. & Zhang, Q. Recent progress in covalent organic frameworks for flexible electronic devices. FlexMat1, 160–172, 10.1002/flm2.26 (2024). 11/15

work page doi:10.1002/flm2.26 2024
[3]

Yuan, S.et al.Covalent organic frameworks for membrane separation.Chem. Soc. Rev.48, 2665–2681, 10.1039/ C8CS00919H (2019)

2019
[4]

& Jiang, D

Guo, J. & Jiang, D. Covalent Organic Frameworks for Heterogeneous Catalysis: Principle, Current Status, and Challenges. ACS Cent. Sci.6, 869–879, 10.1021/acscentsci.0c00463 (2020)

work page doi:10.1021/acscentsci.0c00463 2020
[5]

& Brédas, J.-L

Ni, X., Li, H., Liu, F. & Brédas, J.-L. Engineering of flat bands and Dirac bands in two-dimensional covalent organic frame- works (COFs): Relationships among molecular orbital symmetry, lattice symmetry, and electronic-structure characteristics. Mater. Horizons9, 88–98, 10.1039/D1MH00935D (2022)

work page doi:10.1039/d1mh00935d 2022
[6]

M., Ortega-Guerrero, A

Mourino, B., Jablonka, K. M., Ortega-Guerrero, A. & Smit, B. In Search of Covalent Organic Framework Photocatalysts: A DFT-Based Screening Approach.Adv. Funct. Mater.33, 2301594, 10.1002/adfm.202301594 (2023)

work page doi:10.1002/adfm.202301594 2023
[7]

& Solimannejad, M

Heidari, M. & Solimannejad, M. A DFT study of COF-1 covalent organic framework as a disposable platform for rechargeable lithium-ion battery anodes.Phys. B: Condens. Matter664, 415027, 10.1016/j.physb.2023.415027 (2023)

work page doi:10.1016/j.physb.2023.415027 2023
[8]

& Heine, T

Lukose, B., Kuc, A., Frenzel, J. & Heine, T. On the reticular construction concept of covalent organic frameworks.Beilstein J. Nanotechnol.1, 60–70, 10.3762/bjnano.1.8 (2010)

work page doi:10.3762/bjnano.1.8 2010
[9]

& Cuniberti, G

Raptakis, A., Dianat, A., Croy, A. & Cuniberti, G. Predicting the bulk modulus of single-layer covalent organic frameworks with square-lattice topology from molecular building-block properties.Nanoscale13, 1077–1085, 10.1039/D0NR07666J (2021)

work page doi:10.1039/d0nr07666j 2021
[10]

& Cuniberti, G

Bodesheim, D., Biele, R. & Cuniberti, G. Hierarchies of Hofstadter butterflies in 2D covalent organic frameworks.npj 2D Mater. Appl.7, 16, 10.1038/s41699-023-00378-0 (2023)

work page doi:10.1038/s41699-023-00378-0 2023
[11]

& Ortmann, F

Bittner, E.-A., Merkel, K. & Ortmann, F. Engineering the electrostatic potential in a cof’s pore by selecting quadrupolar building blocks and linkages.npj 2D Mater. Appl.8, 58 (2024)

2024
[12]

& Farzad, F

Ghahari, A., Raissi, H., Pasban, S. & Farzad, F. Proposing two-dimensional covalent organic frameworks material for the capture of phenol molecules from wastewaters.npj Clean Water5, 28, 10.1038/s41545-022-00173-2 (2022)

work page doi:10.1038/s41545-022-00173-2 2022
[13]

P., Galvelis, R

Dürholt, J. P., Galvelis, R. & Schmid, R. Coarse graining of force fields for metal–organic frameworks.Dalton Transactions 45, 4370–4379, 10.1039/C5DT03865K (2016)

work page doi:10.1039/c5dt03865k 2016
[14]

Mohamed, A. M. O., Economou, I. G. & Jeong, H.-K. Coarse-grained force field for ZIF-8: A study on adsorption, diffusion, and structural properties.The J. Chem. Phys.160, 204706, 10.1063/5.0202961 (2024)

work page doi:10.1063/5.0202961 2024
[15]

Alvares, C. M. S., Maurin, G. & Semino, R. Coarse-grained modeling of zeolitic imidazolate framework-8 using MARTINI force fields.The J. Chem. Phys.158, 194107, 10.1063/5.0145924 (2023)

work page doi:10.1063/5.0145924 2023
[16]

& Cuniberti, G

Bodesheim, D., Croy, A. & Cuniberti, G. Elastic Properties of Defective 2D Polymers from Regression Driven Coarse- Graining.J. Chem. Theory Comput.10.1021/acs.jctc.5c01339 (2025)

work page doi:10.1021/acs.jctc.5c01339 2025
[17]

W., Brommer, D

Cranford, S. W., Brommer, D. B. & Buehler, M. J. Extended graphynes: Simple scaling laws for stiffness, strength and fracture.Nanoscale4, 7797–7809, 10.1039/C2NR31644G (2012)

work page doi:10.1039/c2nr31644g 2012
[18]

& Cuniberti, G

Croy, A., Raptakis, A., Bodesheim, D., Dianat, A. & Cuniberti, G. Toward Coarse-Grained Elasticity of Single-Layer Covalent Organic Frameworks.The J. Phys. Chem. C126, 18943–18951, 10.1021/acs.jpcc.2c06268 (2022)

work page doi:10.1021/acs.jpcc.2c06268 2022
[19]

R., Dhakshinamoorthy, A

Daliran, S., Oveisi, A. R., Dhakshinamoorthy, A. & Garcia, H. Probing Defects in Covalent Organic Frameworks.ACS Appl. Mater. & Interfaces16, 50096–50114, 10.1021/acsami.4c12069 (2024)

work page doi:10.1021/acsami.4c12069 2024
[20]

& Mobin, S

Prakash, K., Deka, R. & Mobin, S. M. A review on covalent organic frameworks: Exploration of their growing potential as porous materials in photocatalytic applications.Inorg. Chem. Front.11, 6711–6752, 10.1039/D4QI01480D (2024)

work page doi:10.1039/d4qi01480d 2024
[21]

A.et al.Phonon scattering effects from point and extended defects on thermal conductivity studied via ion irradiation of crystals with self-impurities.Phys

Scott, E. A.et al.Phonon scattering effects from point and extended defects on thermal conductivity studied via ion irradiation of crystals with self-impurities.Phys. Rev. Mater.2, 095001, 10.1103/PhysRevMaterials.2.095001 (2018)

work page doi:10.1103/physrevmaterials.2.095001 2018
[22]

Polanco, C. A. & Lindsay, L. Ab initio phonon point defect scattering and thermal transport in graphene.Phys. Rev. B97, 014303, 10.1103/PhysRevB.97.014303 (2018)

work page doi:10.1103/physrevb.97.014303 2018
[23]

G., Ang, K.-W

Sorkin, V ., Zhou, H., Yu, Z. G., Ang, K.-W. & Zhang, Y .-W. The effects of point defect type, location, and density on the Schottky barrier height of Au/MoS2 heterojunction: A first-principles study.Sci. Reports12, 18001, 10.1038/ s41598-022-22913-7 (2022)

2022
[24]

Liu, W., Han, Y ., Liu, M., Chen, L. & Xu, J. Effect of defects on optical and electronic properties of graphene quantum dots: A density functional theory study.RSC Adv.13, 16232–16240, 10.1039/D3RA02564K (2023)

work page doi:10.1039/d3ra02564k 2023
[25]

Kunstmann, J., Wendumu, T. B. & Seifert, G. Localized defect states in MoS$_2$ monolayers: Electronic and optical properties.physica status solidi (b)254, 1600645, 10.1002/pssb.201600645 (2017). 1606.00347. 12/15

work page doi:10.1002/pssb.201600645 2017
[26]

ACS Mater

Du, K.et al.Impact of Structural Defects on the Electronic Properties of Two-Dimensional Covalent Organic Frameworks. ACS Mater. Lett.6, 335–344, 10.1021/acsmaterialslett.3c01429 (2024)

work page doi:10.1021/acsmaterialslett.3c01429 2024
[27]

& Kumar, S

Thakolkaran, P., Zheng, Y ., Guo, Y ., Vashisth, A. & Kumar, S. Deep learning reveals key predictors of thermal conductivity in covalent organic frameworks, 10.48550/arXiv.2409.06457 (2025). 2409.06457

work page doi:10.48550/arxiv.2409.06457 2025
[28]

S.et al.Epitaxial Ge0.81Sn0.19Nanowires for Nanoscale Mid-Infrared Emitters.ACS Nano13, 8047– 8054 (2019)

Kwon, J.et al.Thermal Conductivity of Covalent-Organic Frameworks.ACS Nano17, 15222–15230, 10.1021/acsnano. 3c03518 (2023)

work page doi:10.1021/acsnano 2023
[29]

Mater.33, 2102807, 10.1002/adma.202102807 (2021)

Mortazavi, B.et al.First-Principles Multiscale Modeling of Mechanical Properties in Graphene/Borophene Heterostructures Empowered by Machine-Learning Interatomic Potentials.Adv. Mater.33, 2102807, 10.1002/adma.202102807 (2021)

work page doi:10.1002/adma.202102807 2021
[30]

Horizons7, 2359–2367, 10.1039/D0MH00787K (2020)

Mortazavi, B.et al.Machine-learning interatomic potentials enable first-principles multiscale modeling of lattice thermal conductivity in graphene/borophene heterostructures.Mater. Horizons7, 2359–2367, 10.1039/D0MH00787K (2020)

work page doi:10.1039/d0mh00787k 2020
[31]

B.et al.Recent advances in machine learning and coarse-grained potentials for biomolecular simulations

Poma, A. B.et al.Recent advances in machine learning and coarse-grained potentials for biomolecular simulations. Biophys. J.124, 1–17 (2025)

2025
[32]

UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations

Rappe, A. K., Casewit, C. J., Colwell, K. S., Goddard, W. A. I. & Skiff, W. M. UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations.J. Am. Chem. Soc.114, 10024–10035, 10.1021/ja00051a040 (1992)

work page doi:10.1021/ja00051a040 1992
[33]

arXiv preprint arXiv:2604.00069(2026)

Medrano Sandonas, L.et al.Perspective: Towards sustainable exploration of chemical spaces with machine learning. arXiv preprint arXiv:2604.00069(2026)

work page arXiv 2026
[34]

Batatia, D

Batatia, I., Kovács, D. P., Simm, G. N. C., Ortner, C. & Csányi, G. MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields, 10.48550/arXiv.2206.07697 (2023). 2206.07697

work page doi:10.48550/arxiv.2206.07697 2023
[35]

Andreas Bender, Nadine Schneider, Marwin Segler, W

Batzner, S.et al.E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials.Nat. Commun.13, 2453, 10.1038/s41467-022-29939-5 (2022)

work page doi:10.1038/s41467-022-29939-5 2022
[36]

& Drautz, R

Bochkarev, A., Lysogorskiy, Y . & Drautz, R. Graph Atomic Cluster Expansion for Semilocal Interactions beyond Equivariant Message Passing.Phys. Rev. X14, 021036, 10.1103/PhysRevX.14.021036 (2024)

work page doi:10.1103/physrevx.14.021036 2024
[37]

T., Unke, O

Frank, J. T., Unke, O. T. & Müller, K.-R. So3krates: Equivariant attention for interactions on arbitrary length-scales in molecular systems, 10.48550/arXiv.2205.14276 (2023). 2205.14276

work page doi:10.48550/arxiv.2205.14276 2023
[38]

Learning local equivariant representations for large-scale atomistic dynamics,

Musaelian, A.et al.Learning local equivariant representations for large-scale atomistic dynamics.Nat. Commun.14, 579, 10.1038/s41467-023-36329-y (2023)

work page doi:10.1038/s41467-023-36329-y 2023
[39]

Gaussian

Bartók, A. P., Payne, M. C., Kondor, R. & Csányi, G. Gaussian Approximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons.Phys. Rev. Lett.104, 136403, 10.1103/PhysRevLett.104.136403 (2010)

work page doi:10.1103/physrevlett.104.136403 2010
[40]

Shapeev, A. V . Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Potentials.Multiscale Model. & Simul.14, 1153–1173, 10.1137/15M1054183 (2016)

work page doi:10.1137/15m1054183 2016
[41]

M.et al.Machine learned potential for high-throughput phonon calculations of metal—organic frameworks.npj Comput

Elena, A. M.et al.Machine learned potential for high-throughput phonon calculations of metal—organic frameworks.npj Comput. Mater.11, 1–12, 10.1038/s41524-025-01611-8 (2025)

work page doi:10.1038/s41524-025-01611-8 2025
[42]

P.et al.MACE-OFF: Short-Range Transferable Machine Learning Force Fields for Organic Molecules.J

Kovács, D. P.et al.MACE-OFF: Short-Range Transferable Machine Learning Force Fields for Organic Molecules.J. Am. Chem. Soc.147, 17598–17611, 10.1021/jacs.4c07099 (2025)

work page doi:10.1021/jacs.4c07099 2025
[43]

Wieser, S., Cen, Y .-J., Madsen, G. K. H. & Carrete, J. Accelerating First-Principles Molecular-Dynamics Thermal Conductivity Calculations for Complex Systems.J. Chem. Theory Comput.22, 513–527, 10.1021/acs.jctc.5c01525 (2026). 44.Liu, Y .et al.Medium-Range Structural Order in Amorphous Arsenic.J. Am. Chem. Soc.148, 9400–9412 (2026)

work page doi:10.1021/acs.jctc.5c01525 2026
[45]

& Zhuang, X

Mortazavi, B., Rabczuk, T. & Zhuang, X. Exploring the structural stability, thermal and mechanical properties of nanoporous carbon nitride nanosheets using a transferrable machine learning interatomic potential.Mach. Learn. for Comput. Sci. Eng.1, 5, 10.1007/s44379-024-00008-6 (2024)

work page doi:10.1007/s44379-024-00008-6 2024
[46]

Batatia, I.et al.A foundation model for atomistic materials chemistry.The J. Chem. Phys.163, 184110, 10.1063/5.0297006 (2025)

work page doi:10.1063/5.0297006 2025
[47]

Ding, S.-Y .et al.Construction of Covalent Organic Framework for Catalysis: Pd/COF-LZU1 in Suzuki–Miyaura Coupling Reaction.J. Am. Chem. Soc.133, 19816–19822, 10.1021/ja206846p (2011). 13/15

work page doi:10.1021/ja206846p 2011
[48]

Guo, H.et al.Catalyst-derived hierarchy in 2D imine-based covalent organic frameworks.Nanoscale17, 6488–6504, 10.1039/D4NR04717F (2025)

work page doi:10.1039/d4nr04717f 2025
[49]

Li, X.et al.Construction of functional covalent organic framework films by modulator and solvent induced polymerization. Nat. Commun.16, 1223, 10.1038/s41467-024-55114-z (2025)

work page doi:10.1038/s41467-024-55114-z 2025
[50]

& Ayesh, Ahmad

Majidi, R. & Ayesh, Ahmad. I. First-principles investigation of electronic and optical properties of COF-LZU1: A two-dimensional covalent organic framework.Comput. Theor. Chem.1254, 115493, 10.1016/j.comptc.2025.115493 (2025)

work page doi:10.1016/j.comptc.2025.115493 2025
[51]

Liang, B.et al.Optimal acceleration voltage for near-atomic resolution imaging of layer-stacked 2D polymer thin films. Nat. Commun.13, 3948, 10.1038/s41467-022-31688-4 (2022)

work page doi:10.1038/s41467-022-31688-4 2022
[52]

Tian, Z., Esfarjani, K., Shiomi, J., Henry, A. S. & Chen, G. On the importance of optical phonons to thermal conductivity in nanostructures.Appl. Phys. Lett.99, 053122, 10.1063/1.3615709 (2011)

work page doi:10.1063/1.3615709 2011
[53]

Sci.16, 3738–3754, 10.1039/D4SC06530A (2025)

Poltavsky, I.et al.Crash testing machine learning force fields for molecules, materials, and interfaces: Molecular dynamics in the TEA challenge 2023.Chem. Sci.16, 3738–3754, 10.1039/D4SC06530A (2025)

work page doi:10.1039/d4sc06530a 2023
[54]

Ouyang, T.et al.First-principles study of thermal transport in nitrogenated holey graphene.Nanotechnology28, 045709, 10.1088/1361-6528/28/4/045709 (2016)

work page doi:10.1088/1361-6528/28/4/045709 2016
[55]

A review on brain tumor segmentation based on deep learning methods with federated learning techniques

Arabha, S. & Rajabpour, A. Thermo-mechanical properties of nitrogenated holey graphene (C2N): A comparison of machine-learning-based and classical interatomic potentials.Int. J. Heat Mass Transf.178, 121589, 10.1016/j. ijheatmasstransfer.2021.121589 (2021). 2105.08940

work page doi:10.1016/j 2021
[56]

K., Phillpot, S

Schelling, P. K., Phillpot, S. R. & Keblinski, P. Comparison of atomic-level simulation methods for computing thermal conductivity.Phys. Rev. B65, 144306, 10.1103/PhysRevB.65.144306 (2002)

work page doi:10.1103/physrevb.65.144306 2002
[57]

van Duin, A. C. T., Dasgupta, S., Lorant, F. & Goddard, W. A. ReaxFF: A Reactive Force Field for Hydrocarbons.The J. Phys. Chem. A105, 9396–9409, 10.1021/jp004368u (2001)

work page doi:10.1021/jp004368u 2001
[58]

Chenoweth, K., van Duin, A. C. T. & Goddard, W. A. ReaxFF Reactive Force Field for Molecular Dynamics Simulations of Hydrocarbon Oxidation.The J. Phys. Chem. A112, 1040–1053, 10.1021/jp709896w (2008). 60.Landau, L. D., Pitaevskii, L. P., Kosevich, A. M. & Lifshitz, E. M.Theory of Elasticity: V olume 7(Elsevier, 2012)

work page doi:10.1021/jp709896w 2008
[59]

& Grossman, J

Lin, L.-C., Choi, J. & Grossman, J. C. Two-dimensional covalent triazine framework as an ultrathin-film nanoporous membrane for desalination.Chem. Commun.51, 14921–14924, 10.1039/C5CC05969K (2015)

work page doi:10.1039/c5cc05969k 2015
[60]

& Zhang, J

Wang, B., Ying, P. & Zhang, J. The thermoelastic properties of monolayer covalent organic frameworks studied by machine-learning molecular dynamics.Nanoscale16, 237–248, 10.1039/D3NR04509A (2023)

work page doi:10.1039/d3nr04509a 2023
[61]

Castano, I.et al.Mapping Grains, Boundaries, and Defects in 2D Covalent Organic Framework Thin Films.Chem. Mater. 33, 1341–1352, 10.1021/acs.chemmater.0c04382 (2021)

work page doi:10.1021/acs.chemmater.0c04382 2021
[62]

Haase, F.et al.Tuning the stacking behaviour of a 2d covalent organic framework through non-covalent interactions. Mater . Chem. Front.1, 1354–1361 (2017)

2017
[63]

De Bolòs, E.et al.Interlocked 2d covalent organic frameworks from overcrowded nodes.J. Am. Chem. Soc.147, 2579–2586, 10.1021/jacs.4c14453 (2025). PMID: 39803845, https://doi.org/10.1021/jacs.4c14453

work page doi:10.1021/jacs.4c14453 2025
[64]

Ikeda, Y .et al.Machine-Learning Interatomic Potentials Achieving CCSD(T) Accuracy for Systems with Extended Covalent Networks and van der Waals Interactions.J. Chem. Theory Comput.22, 2739–2756, 10.1021/acs.jctc.5c02045 (2026)

work page doi:10.1021/acs.jctc.5c02045 2026
[65]

Gardner, J. L. A.et al.Distillation of atomistic foundation models across architectures and chemical domains. https://arxiv.org/abs/2506.10956v1 (2025)

work page arXiv 2025
[66]

Larsen and Jens Jørgen Mortensen and Jakob Blomqvist and Ivano E

Hjorth Larsen, A.et al.The atomic simulation environment—a Python library for working with atoms.J. Physics: Condens. Matter29, 273002, 10.1088/1361-648X/aa680e (2017)

work page doi:10.1088/1361-648x/aa680e 2017
[67]

and Aktulga, H

Thompson, A. P.et al.LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales.Comput. Phys. Commun.271, 108171, 10.1016/j.cpc.2021.108171 (2022)

work page doi:10.1016/j.cpc.2021.108171 2021
[68]

& Cuniberti, G

Medrano Sandonas, L., Gutierrez, R., Dianat, A. & Cuniberti, G. Engineering thermal rectification in MoS2 nanoribbons: A non-equilibrium molecular dynamics study.RSC Adv.5, 54345–54351, 10.1039/C5RA05733G (2015)

work page doi:10.1039/c5ra05733g 2015
[69]

Medrano Sandonas, L.et al.Enhancement of thermal transport properties of asymmetric Graphene/hBN nanoribbon heterojunctions by substrate engineering.Carbon124, 642–650, 10.1016/j.carbon.2017.09.025 (2017). 14/15

work page doi:10.1016/j.carbon.2017.09.025 2017
[70]

Data-Driven Characterization of (A)Chiral 2D Polymers

Ikeshoji, T. & Hafskjold, B. Non-equilibrium molecular dynamics calculation of heat conduction in liquid and through liquid-gas interface.Mol. Phys.81, 251–261, 10.1080/00268979400100171 (1994). Acknowledgements A.S. and G.C. acknowledge funding by the DFG project "Data-Driven Characterization of (A)Chiral 2D Polymers" (CRC1415 - C04) L.M.S and G.C. grate...

work page doi:10.1080/00268979400100171 1994