With the rise of large-scale simulation, machine learning and other cutting-edge applications, distributed computing has become an important means of computing research. In this work, we apply coding computing to the field of fragmented quantum chemistry: On the one hand, by using gradient coding scheme for reference, we solve the problem of lagging nodes in distributed computing; On the other hand, the automatic error correction ability of distributed computing is increased to reduce the cost of human and material resources in the calculation process, so as to realize automatic fault-tolerant quantum chemical calculation. In addition, we also put forward the idea of coding reuse and QM/MM hierarchical computing, which can simply and effectively use more computing resources to carry out distributed computing on the set fault tolerance. 

This work was published in Acta Chimica Sinica (JCR Q3). The first author is Li Ning, a joint graduate student of Wenzhou University and Computer Network Information Center (CNIC). The corresponding authors are associate Professor Ma Yingjin of CNIC, Associate Professor Fang Guoyong of Wenzhou University, separately.

“Encode-compute-decode” fault-tolrent distributed quantum chemical calculation

At the meantime, we combined this calculation scheme with machine learning-assisted shard load prediction, dynamic and static load balancing, renormalization excited state calculation and other methods to initially realize intelligent fault-tolerant large-system shard calculation. Finally, we apply this scheme to the calculation of binding energy of P38 protein and ligand, and the calculation of cluster systems with single or multi-center excitation, and compare the results obtained by coding calculation with the real results to verify the accuracy of this scheme and its application potential in automated fault-tolerant quantum chemistry calculation. 

This work accepted by Journal of Computational Chemistry as the invited research paper, the co-first authors are Yuan Kai of Beijing University of Chemical Technology (BUCT), Zhou Shuai of Institute of Computing Technology, the corresponding authors are associate Professor Ma Yingjin of CNIC, Professor Guo Danhuai of BUCT, separately.

These works were supported by National Natural Science Foundation of China, Strategic Priority Research Program of Chinese Academy of Sciences, Youth Innovation Promotion Association of Chinese Academy of Sciences, Network and Information Foundation of Chinese Academy of Sciences, and Project of Computer Network Information Center.

Machine-Learning assisted Fault-tolerant distributed quantum chemical calculations

 References:

1. Fault-tolerant Coded Quantum Chemical Distributed Calculation，Ning Li, Lina Xu, Guoyong Fang*, Yingjin Ma*, Acta Chimica Sinica, 82, 138-145 (2024) https://sioc-journal.cn/Jwk_hxxb/EN/10.6023/A23110496

2. Fault-tolerant Quantum Chemical Calculations with Improved Machine-Learning Models, Kai Yuan, Shuai Zhou, Ning Li, Tianyan Li, Danhuai Guo*, Yingjin Ma*, J. Comput. Chem. accepted. https://onlinelibrary.wiley.com/journal/1096987x