Global optimization of crystal compositions is a significant yet computationally intensive method to identify stable structures within chemical space.The specific physical properties linked to a threedimensional atomi...Global optimization of crystal compositions is a significant yet computationally intensive method to identify stable structures within chemical space.The specific physical properties linked to a threedimensional atomic arrangement make this an essential task in the development of new materials.We present a method that efficiently uses active learning of neural network force fields for structure relaxation,minimizing the required number of steps in the process.This is achieved by neural network force fields equipped with uncertainty estimation,which iteratively guide a pool of randomly generated candidates toward their respective local minima.Using this approach,we are able to effectively identify themost promising candidates for further evaluation using density functional theory(DFT).Our method not only reliably reduces computational costs by up to two orders of magnitude across the benchmark systemsSi_(16),Na_(8)Cl_(8),Ga_(8)As_(8)and Al_(4)O_(6)but also excels in finding themost stable minimum for the unseen,more complex systems Si46 and Al16O24.Moreover,we demonstrate at the example of Si_(16)that our method can find multiple relevant local minima while only adding minor computational effort.展开更多
基金N.W.A.G.and M.G.contributed to this research while working at the BASLEARN-TU Berlin/BASF Joint Lab for Machine Learning,co-financed by TU Berlin and BASF SE.K.T.S.contributed to this research while working at TU Berlin and BIFOLD with grant number 01IS18037Asupported by JSPS KAKENHI Grant Number JP23H05457 and by JST-CREST Grant Number JPMJCR22O2.We thank Jonas Lederer and Klaus-Robert Müller for insightful discussions and feedback.
文摘Global optimization of crystal compositions is a significant yet computationally intensive method to identify stable structures within chemical space.The specific physical properties linked to a threedimensional atomic arrangement make this an essential task in the development of new materials.We present a method that efficiently uses active learning of neural network force fields for structure relaxation,minimizing the required number of steps in the process.This is achieved by neural network force fields equipped with uncertainty estimation,which iteratively guide a pool of randomly generated candidates toward their respective local minima.Using this approach,we are able to effectively identify themost promising candidates for further evaluation using density functional theory(DFT).Our method not only reliably reduces computational costs by up to two orders of magnitude across the benchmark systemsSi_(16),Na_(8)Cl_(8),Ga_(8)As_(8)and Al_(4)O_(6)but also excels in finding themost stable minimum for the unseen,more complex systems Si46 and Al16O24.Moreover,we demonstrate at the example of Si_(16)that our method can find multiple relevant local minima while only adding minor computational effort.
