Force-based gradient descent method for ab initio atomic structure relaxation

作     者：Yukuan Hu Xingyu Gao Yafan Zhao Xin Liu Haifeng Song 

作者机构：State Key Laboratory of Scientific and Engineering Computing Academy of Mathematics and Systems Science Chinese Academy of Sciences Beijing 100190 China University of Chinese Academy of Sciences Beijing 100049 China Laboratory of Computational Physics Institute of Applied Physics and Computational Mathematics Beijing 100088 China CAEP Software Center for High Performance Numerical Simulation Beijing 100088 China 

出 版 物：《Physical Review B》 (Phys. Rev. B)

年 卷 期：2022年第106卷第10期

页      面：104101-104101页

核心收录：

基　　金：Chinese Academy of Sciences (CAS) AMSS-PolyU Joint Laboratory in Applied Mathematics National Natural Science Foundation of China, NSFC, (11971466, 11991020, 11991021, 12021001, 12125108, 12288201) National Natural Science Foundation of China, NSFC Chinese Academy of Sciences, CAS, (ZDBS-LY-7022) Chinese Academy of Sciences, CAS Science Challenge Project, (TZ2018002) Science Challenge Project 

主　　题：Crystal structure Structural properties Density functional theory development 

摘      要：Force-based algorithms for ab initio atomic structure relaxation, such as conjugate gradient methods, usually get stuck in the line minimization processes along search directions, where expensive ab initio calculations are triggered frequently to test trial positions before locating the next iterate. We present a force-based gradient descent method, WANBB, that circumvents the deficiency. At each iteration, WANBB enters the line minimization process with a trial step size capturing the local curvature of the energy surface. The exit is controlled by a nonrestrictive criterion that tends to accept early trials. These two ingredients streamline the line minimization process in WANBB. The numerical simulations on nearly 80 systems with good universality demonstrate the considerable compression of WANBB on the cost for the unaccepted trials compared with conjugate gradient methods. We also observe across the board significant and universal speedups as well as the superior robustness of WANBB over several widely used methods. The latter point is theoretically established. The implementation of WANBB is pretty simple, in that no a priori physical knowledge is required and only three parameters are present without tuning.