Pushing the limit of molecular dynamics with ab initio accuracy 100 million atoms with machine learning

W Jia, H Wang, M Chen, D Lu, J Liu, L Lin, R Car, W. E, L. Zhang
arXiv preprint arXiv:2005.00223 (2020)
Submitted to the supercomputing conference SC20
Atlanta, Nov15, 20

Abstract
For 35 years, {\it ab initio} molecular dynamics (AIMD) has been the method of choice for understanding complex materials and molecules at the atomic scale from first principles. However, most applications of AIMD are limited to systems with thousands of atoms due to the high computational complexity. We report that a machine learning-based molecular simulation protocol (Deep Potential Molecular Dynamics), driven by a highly optimized code (GPU DeePMD-kit) on the Summit supercomputer, has greatly expanded the capabilities of MD simulation with {\it ab initio} accuracy, pushing its limit to simulation of over 100 million atoms for one nanosecond per day. Our code can efficiently scale up to the entire Summit supercomputer, reaching 86 PFLOPS in double precision (43% of the peak) and 137 PFLOPS in mixed precision. This success opens the door to the modeling of atomic processes in realistic materials and molecular systems with {\it ab initio} accuracy.