Pretraining of attention-based deep learning potential model for molecular simulation

作     者：Duo Zhang Hangrui Bi Fu-Zhi Dai Wanrun Jiang Xinzijian Liu Linfeng Zhang Han Wang 

作者机构：AI for Science InstituteBeijing 100080China DP TechnologyBeijing 100080China Academy for Advanced Interdisciplinary StudiesPeking UniversityBeijing 100871China Laboratory of Computational PhysicsInstitute of Applied Physics and Computational MathematicsBeijing 100094China HEDPSCAPTCollege of EngineeringPeking UniversityBeijing 100871China 

出 版 物：《npj Computational Materials》 (计算材料学(英文))

年 卷 期：2024年第10卷第1期

页      面：2297-2304页

核心收录：

学科分类：07[理学] 0701[理学-数学] 070101[理学-基础数学] 

基　　金：supported by the National Key R&D Program of China under Grant No.2022YFA1004300 the National Natural Science Foundation of China under Grant No.12122103 supported by the Bohrium Cloud Platform at DP technology 

主　　题：potential representing positions 

摘      要：Machine learning-assisted modeling of the inter-atomic potential energy surface(PES)is revolutionizing the field ofmolecular *** the accumulation of high-quality electronic structure data,a model that can be pretrained on all available data and finetuned on downstream tasks with a small additional effort would bring the field to a new *** we propose DPA-1,a Deep Potentialmodel with a gated attentionmechanism,which is highly effective for representing the conformation and chemical spaces of atomic systems and learning the *** tested DPA-1 on a number of systems and observed superior performance compared with existing *** pretrained on large-scale datasets containing 56 elements,DPA-1 can be successfully applied to various downstream tasks with a great improvement of sample ***,for different elements,the learned type embedding parameters form a spiral in the latent space and have a natural correspondence with their positions on the periodic table,showing interesting interpretability of the pretrained DPA-1 model.