Report title: The Latest Development of the In-depth Potential Energy Method
Reporter: Wang Han, researcher, Beijing Institute of Applied Physics and Computational Mathematics
Reporting time: 8:50-9:25 am, September 17, 2020
Report location: Tencent Conference ID: 206 372 412
Conference password: 0917
School contact: Jia Jiwei jiajiwei@jlu.edu.cn
Report summary:
Material design on electronic computers requires a precise description of the interaction between atoms. The depth potential energy method is equivalent to the first-principles method in accuracy, and the calculation efficiency is equivalent to the traditional empirical force field, so it is a very promising method for modeling atomic interaction. In this report, we introduce the latest developments in deep potential energy methods: (1) Modeling of tensor-type physical quantities. In the model, we guarantee the invariance or covariance of tensor-type physical quantities under symmetric operations. This function allows us to calculate the observables in experiments such as IR spectrum and Raman spectrum with first-principles accuracy, providing a powerful tool for the interpretation of experimental results and the analysis of physical laws. (2) The optimization of deep potential energy in contemporary heterogeneous supercomputers. Our work shows that the organic combination of physical model + deep learning + high-performance computing enables molecular dynamics simulations with first-principles accuracy on the order of billion atoms to be completed within a day. This work has improved the existing first-principles precision molecular simulation capabilities by more than three orders of magnitude, opening up new possibilities for molecular simulations. (3) The post-depth Hartree-Fock method, a new method that uses Hartree-Fock/density functional theory calculation cost for high-order quantization correction, provides the possibility for high-precision energy functional modeling.
Brief introduction of the speaker:
Wang Han is a researcher at the Beijing Institute of Applied Physics and Computational Mathematics. He received his undergraduate degree in Computational Mathematics from Peking University in 2006 and a PhD in Computational Mathematics from Peking University in 2011. During 2007-2008, He Visited and studied at Max Planck Institute in Germany. From 2011 to 2014, he was engaged in post-doctoral research at Freie Universität Berlin in Germany. From 2014 to 2018, he worked as a research scientist in the High Performance Numerical Simulation Software Center of China Academy of Engineering Physics. Since 2018, he has been a research scientist at the Beijing Institute of Applied Physics and Computational Mathematics.
Researcher Wang Han has long been engaged in molecular dynamics numerical analysis and fast algorithm research, as well as multi-scale model and simulation research, and published more than 30 academic papers. The molecular dynamics-atom interaction potential modeling project based on deep learning developed by researcher Wang Han and his team has received wide attention and recognition in the professional field, and successfully released the open source software DeePMD-kit, which has important impact to the field.
