Authors: Honghui Shang (State Key Laboratory of Precision and Intelligent Chemistry, Hefei National Laboratory, University of Science and Technology of China); Zhiyu Xiao (Institute of Physics, Chinese Academy of Sciences); Zhuozhao Xia, Hao Wu, and Huan Ma (State Key Laboratory of Precision and Intelligent Chemistry, Hefei National Laboratory, University of Science and Technology of China); Yongqing Chen and Bin Li (National High Performance Computer Engineering Technology Research Center); Zhenyu Li (State Key Laboratory of Precision and Intelligent Chemistry, Hefei National Laboratory, University of Science and Technology of China); Yutong Lu (Sun Yat-sen University); and Jinlong Yang (State Key Laboratory of Precision and Intelligent Chemistry, Hefei National Laboratory, University of Science and Technology of China)
Abstract: The FeMo-cofactor (FeMoco) of nitrogenase has for nearly a decade been the canonical "killer app" motivating fault-tolerant quantum computing for chemistry: the CAS(113e, 76o) active space is beyond exact diagonalization and remains challenging for every polynomial-scaling classical method. We report the first exascale classical solution of this active space. The resulting code has been tuned for two exascale supercomputers, LineShine and China's New-generation Intelligent Supercomputer (CNIS), reaching sustained performance on 20,480 LineShine nodes of 1.456 EFLOP/s in double precision (58.9% of the machine peak) and 5.626 EFLOP/s in mixed precision. These computational achievements enable the FeMoco ground-state consensus to be reached on 1,280 LineShine nodes within a measured end-to-end wall time of 105 minutes (1.75 h), approximately one-fifth of the best published 8.6-hour single-run projection for fault-tolerant quantum computing on the same active space.