There has been an ongoing race for the past several years to develop the best universal machine learning interatomic potential.This progress has led to increasingly accurate models for predicting energy,forces,and str...There has been an ongoing race for the past several years to develop the best universal machine learning interatomic potential.This progress has led to increasingly accurate models for predicting energy,forces,and stresses,combining innovative architectures with big data.Here,we benchmark these models on their ability to predict harmonic phonon properties,which are critical for understanding the vibrational and thermal behavior of materials.Using around 10000 ab initio phonon calculations,we evaluate model performance across various phonon-related parameters to test the universal applicability of these models.The results reveal that some models achieve high accuracy in predicting harmonic phonon properties.However,others still exhibit substantial inaccuracies,even if they excel in the prediction of the energy and the forces for materials close to dynamical equilibrium.These findings highlight the importance of considering phonon-related properties in the development of universal machine learning interatomic potentials.展开更多
Artificial neural network(ANN)potentials enable accurate atomistic simulations of complex materials at unprecedented scales,but training them for potential energy surfaces(PES)of diverse chemical environments remains ...Artificial neural network(ANN)potentials enable accurate atomistic simulations of complex materials at unprecedented scales,but training them for potential energy surfaces(PES)of diverse chemical environments remains computationally intensive,especially when the PES gradients are trained on atomic force data.Here,we present an efficient methodology incorporating forces intoANNtraining by translating them to synthetic energy data using Gaussian process regression(GPR),leading to accurate PES models with fewer additional first-principles calculations and a reduced computational effort for training.We evaluated the method on hybrid density-functional theory data for ethylene carbonate(EC)molecules and their interfaces with Li metal,which are relevant for Li-metal batteries.The GPR-ANN potentials achieved an accuracy comparable to fully force-trained ANN potentials with a significantly reduced computational and memory overhead,establishing the method as a powerful and scalable framework for constructing high-fidelity ANN potentials for complex materials systems.展开更多
基金funding from the Horizon Europe MSCA Doctoral network grant n.101073486, EUSpecLab, funded by the European UnionS.B. and D.S. acknowledge financial support from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the project BO4280/11-1. H.C.W and M.A.L.M would like to thank the NHR Center PC2 for providing computing time on the Noctua 2 supercomputers.
文摘There has been an ongoing race for the past several years to develop the best universal machine learning interatomic potential.This progress has led to increasingly accurate models for predicting energy,forces,and stresses,combining innovative architectures with big data.Here,we benchmark these models on their ability to predict harmonic phonon properties,which are critical for understanding the vibrational and thermal behavior of materials.Using around 10000 ab initio phonon calculations,we evaluate model performance across various phonon-related parameters to test the universal applicability of these models.The results reveal that some models achieve high accuracy in predicting harmonic phonon properties.However,others still exhibit substantial inaccuracies,even if they excel in the prediction of the energy and the forces for materials close to dynamical equilibrium.These findings highlight the importance of considering phonon-related properties in the development of universal machine learning interatomic potentials.
基金support by the Columbia Center for Computational Electrochemistry(CCCE)and computing resources from Columbia University’s Shared Research Computing FacilityJ.L.Z.and N.A.thank the Project HPC-EUROPA3(Grant No.INFRAIA-2016-1-730897)for its support,provided through the EC Research and Innovation Action under the H2020 ProgramN.A.also acknowledges support by a start-up grant(Dutch Sector Plan)from Utrecht University.The authors gratefully acknowledge discussions with JoséA.Garrido Torres and Thomas Bligaard.
文摘Artificial neural network(ANN)potentials enable accurate atomistic simulations of complex materials at unprecedented scales,but training them for potential energy surfaces(PES)of diverse chemical environments remains computationally intensive,especially when the PES gradients are trained on atomic force data.Here,we present an efficient methodology incorporating forces intoANNtraining by translating them to synthetic energy data using Gaussian process regression(GPR),leading to accurate PES models with fewer additional first-principles calculations and a reduced computational effort for training.We evaluated the method on hybrid density-functional theory data for ethylene carbonate(EC)molecules and their interfaces with Li metal,which are relevant for Li-metal batteries.The GPR-ANN potentials achieved an accuracy comparable to fully force-trained ANN potentials with a significantly reduced computational and memory overhead,establishing the method as a powerful and scalable framework for constructing high-fidelity ANN potentials for complex materials systems.
