High-throughput search for new crystal structures is extensively assisted by data-driven solutions.Here we address their prospects for more narrowly focused applications in a data-efficient manner.To verify and experi...High-throughput search for new crystal structures is extensively assisted by data-driven solutions.Here we address their prospects for more narrowly focused applications in a data-efficient manner.To verify and experimentally validate the proposed approach,we consider the structure of higher tungsten borides,WB_(4.2),and eightmetals asWsubstituents to set a search space comprising 375k+inequivalent crystal structures for solid solutions.Their thermodynamic properties are predicted with errors of a few meV/atom using graph neural networks fine-tuned on the DFT-derived properties of ca.200 entries.Amongthe substituents considered,Ta provides thewidest range of predicted stable concentrations and leads to the most considerable changes inmechanical properties.The vacuumless arc plasmamethod is used to perform synthesis of higher tungsten borides with different concentrations of Ta.Vickers hardness of WB_(5-x)samples with different Ta contents is measured,showing increase in hardness.展开更多
The development of fast ionic conductors to improve the performance of electrochemical devices relies on expensive high-throughput(HT)density functional theory(DFT)calculations of transport properties.Machine learning...The development of fast ionic conductors to improve the performance of electrochemical devices relies on expensive high-throughput(HT)density functional theory(DFT)calculations of transport properties.Machine learning(ML)can accelerate HT workflows but requires high-quality data to ensure accurate predictions from trained models.In this study,we introduce the LiTraj dataset,which comprises 13,000 percolation and 122,000 migration barriers,and 1700 migration trajectories,calculated for Li-ion in diverse crystal structures using empirical force fields and DFT,respectively.With LiTraj,we demonstrate that classicalMLmodels and graph neural networks(GNNs)for structureto-property prediction of percolation and migration barriers can distinguish between“fast”and“poor”ionic conductors.Furthermore,we evaluate the capability of GNN-based universal ML interatomic potentials(uMLIPs)to identify optimal Li-ion migration trajectories.Fine-tuned uMLIPs achieve near-DFT accuracy in predicting migration barriers,significantly accelerating HT screenings of new ionic conductors.展开更多
基金platform(Sber,Moscow,Russia)used for calculations with GNN models.DFT calculations were carried out using Skoltech supercomputer Zhores.Experiments on the vacuumless synthesis of higher tungsten boride were carried out with support from the Ministry of Science,Higher Education of the Russian Federation in part of the Science program(Project FSWW-2025-0003).
文摘High-throughput search for new crystal structures is extensively assisted by data-driven solutions.Here we address their prospects for more narrowly focused applications in a data-efficient manner.To verify and experimentally validate the proposed approach,we consider the structure of higher tungsten borides,WB_(4.2),and eightmetals asWsubstituents to set a search space comprising 375k+inequivalent crystal structures for solid solutions.Their thermodynamic properties are predicted with errors of a few meV/atom using graph neural networks fine-tuned on the DFT-derived properties of ca.200 entries.Amongthe substituents considered,Ta provides thewidest range of predicted stable concentrations and leads to the most considerable changes inmechanical properties.The vacuumless arc plasmamethod is used to perform synthesis of higher tungsten borides with different concentrations of Ta.Vickers hardness of WB_(5-x)samples with different Ta contents is measured,showing increase in hardness.
基金the financial support of Russian Science Foundation project No.24-73-10204.
文摘The development of fast ionic conductors to improve the performance of electrochemical devices relies on expensive high-throughput(HT)density functional theory(DFT)calculations of transport properties.Machine learning(ML)can accelerate HT workflows but requires high-quality data to ensure accurate predictions from trained models.In this study,we introduce the LiTraj dataset,which comprises 13,000 percolation and 122,000 migration barriers,and 1700 migration trajectories,calculated for Li-ion in diverse crystal structures using empirical force fields and DFT,respectively.With LiTraj,we demonstrate that classicalMLmodels and graph neural networks(GNNs)for structureto-property prediction of percolation and migration barriers can distinguish between“fast”and“poor”ionic conductors.Furthermore,we evaluate the capability of GNN-based universal ML interatomic potentials(uMLIPs)to identify optimal Li-ion migration trajectories.Fine-tuned uMLIPs achieve near-DFT accuracy in predicting migration barriers,significantly accelerating HT screenings of new ionic conductors.
