Understanding and accurately predicting hydrogen diffusion in materials is challenging due to the complex interactions between hydrogen defects and the crystal lattice.These interactions span large length and time sca...Understanding and accurately predicting hydrogen diffusion in materials is challenging due to the complex interactions between hydrogen defects and the crystal lattice.These interactions span large length and time scales,making them difficult to address with standard ab-initio techniques.This work addresses this challenge by employing accelerated machine learning(ML)molecular dynamics simulations through active learning.We conduct a comparative study of different ML-based interatomic potential schemes,including VASP,MACE,and CHGNet,utilizing various training strategies such as on-the-fly learning,pre-trained universal models,and fine-tuning.By considering different temperatures and concentration regimes,we obtain hydrogen diffusion coefficients and activation energy values which align remarkably well with experimental results,underlining the efficacy and accuracy of ML-assisted methodologies in the context of diffusive dynamics.Particularly,our procedure significantly reduces the computational effort associated with traditional transition state calculations or ad-hoc designed interatomic potentials.The results highlight the limitations of pre-trained universal solutions for defective materials and how they can be improved by fine-tuning.Specifically,fine-tuning the models on a database produced during on-the-fly training of VASP ML force-field allows the retrieval of DFT-level accuracy at a fraction of the computational cost.展开更多
Conductive polymer nanocomposites have emerged as essential materials for wearable devices.In this study,we propose a novel approach that combines graph attention networks(GAT)with an improved global pooling strategy ...Conductive polymer nanocomposites have emerged as essential materials for wearable devices.In this study,we propose a novel approach that combines graph attention networks(GAT)with an improved global pooling strategy and incremental learning.We train the GAT model on homopolymer/carbon nanotube(CNT)nanocomposite data simulated by hybrid particle-field molecular dynamics(hPF-MD)method within the CNT concentration range of 1–8%.We further analyze the conductive network structure by integrating the resistor network approach with the GAT’s attention scores,revealing optimal connectivity at a 7%concentration.The comparative analysis of trained data and the reconstructed network,based on the attention scores,underscores the GATmodel’s ability in learning network structural representations.This work not only validates the efficacy of the GAT model in property prediction and interpretable network structure analysis of polymer nanocomposites but also lays a cornerstone for the reverse engineering of polymer composites.展开更多
基金the“Doctoral College Advanced Functional Materials-Hierarchical Design of Hybrid Systems DOC 85 doc.funds”funded by the Austrian Science Fund(FWF)and by the Vienna Doctoral School in Physics(VDSP),For Open Access purposes,the author has applied a CC BY public copyright license to any author accepted manuscript version arising from this submission.D.M.and S.P.were supported by the European Union Horizon 2020 research and innovation program under Grant Agreement No.857470the European Regional Development Fund under the program of the Foundation for Polish Science International Research Agenda PLUS,grant No.MAB PLUS/2018/8+2 种基金the initiative of the Ministry of Science and Higher Education’Support for the activities of Centers of Excellence established in Poland under the Horizon 2020 program’under agreement No.MEiN/2023/DIR/3795L.P.and C.F.acknowledge the National Recovery and Resilience Plan(NRRP),Mission 4 Component 2 Investment 1.3-Project NEST(Network 4 Energy Sustainable Transition)of Ministero dell’Universitáe della Ricerca(MUR),funded by the European Union-NextGenerationEUL.L.and C.F.acknowledge the NRRP,CN-HPC grant no.(CUP)J33C22001170001,SPOKE 7,of MUR,funded by the European Union-NextGenerationEU.The computational results were obtained using the Vienna Scientific Cluster(VSC)and the LEONARDO cluster.We acknowledge access to LEONARDO at CINECA,Italy,via an AURELEO(Austrian Users at LEONARDO supercomputer)project.
文摘Understanding and accurately predicting hydrogen diffusion in materials is challenging due to the complex interactions between hydrogen defects and the crystal lattice.These interactions span large length and time scales,making them difficult to address with standard ab-initio techniques.This work addresses this challenge by employing accelerated machine learning(ML)molecular dynamics simulations through active learning.We conduct a comparative study of different ML-based interatomic potential schemes,including VASP,MACE,and CHGNet,utilizing various training strategies such as on-the-fly learning,pre-trained universal models,and fine-tuning.By considering different temperatures and concentration regimes,we obtain hydrogen diffusion coefficients and activation energy values which align remarkably well with experimental results,underlining the efficacy and accuracy of ML-assisted methodologies in the context of diffusive dynamics.Particularly,our procedure significantly reduces the computational effort associated with traditional transition state calculations or ad-hoc designed interatomic potentials.The results highlight the limitations of pre-trained universal solutions for defective materials and how they can be improved by fine-tuning.Specifically,fine-tuning the models on a database produced during on-the-fly training of VASP ML force-field allows the retrieval of DFT-level accuracy at a fraction of the computational cost.
基金support from the National Natural Science Foundation of China(52273019)Fundamental Research Funds for the Central Universities(044420250076)+2 种基金General Program of National Natural Science Foundation of Liao Ning Province(2025-MS-105)Scientific Research Funds Project of Liaoning Provincial Department of Education(LJKZ0034)Beijing Natural Science Foundation(4242040).
文摘Conductive polymer nanocomposites have emerged as essential materials for wearable devices.In this study,we propose a novel approach that combines graph attention networks(GAT)with an improved global pooling strategy and incremental learning.We train the GAT model on homopolymer/carbon nanotube(CNT)nanocomposite data simulated by hybrid particle-field molecular dynamics(hPF-MD)method within the CNT concentration range of 1–8%.We further analyze the conductive network structure by integrating the resistor network approach with the GAT’s attention scores,revealing optimal connectivity at a 7%concentration.The comparative analysis of trained data and the reconstructed network,based on the attention scores,underscores the GATmodel’s ability in learning network structural representations.This work not only validates the efficacy of the GAT model in property prediction and interpretable network structure analysis of polymer nanocomposites but also lays a cornerstone for the reverse engineering of polymer composites.
