生物分子 z-bio 2025-01-08 467

　　Under the support of the National Natural Science Foundation of China (T2222026) and other programs, the research achievement "Pushing the Limit of Quantum Mechanical Simulation to the Raman Spectrum of a Biological System with 100 Million Atoms" completed by Professor Shang Honghui and Professor Yang Jinlong of the Key Laboratory of Precision and Intelligent Chemistry of the University of Science and Technology of China in cooperation with Liu Ying, a senior engineer of the Institute of Computing Technology of the Chinese Academy of Sciences, and Professor He Xiao of East China Normal University was successfully shortlisted for the Gordon Bell Award in 2024（ https://sc24.supercomputing.org/2024/10/presenting-the-finalists-for-the-2024-gordon-bell-prize/ ）. This is the only research achievement from China that has been shortlisted, and it is also the team's second nomination for this award since 2021. The Gordon Bell Prize is the highest international award in the field of high-performance computing applications, awarded by the American Computer Society (ACM) to recognize outstanding achievements in high-performance computing worldwide, especially innovative work in the application of high-performance computing to science, engineering, and large-scale data analysis. It is known as the "Nobel Prize in Supercomputing". Paper link: https://dl.acm.org/doi/10.1109/SC41406.2024.00011 .

　　Raman spectroscopy is an important tool for studying the structure of biomolecules, widely used in fields such as drug development and disease diagnosis. However, the quantum simulation of Raman spectroscopy requires enormous computational resources. The previous Raman spectrum quantum simulation can only deal with small systems with thousands of atoms. The QF-RAMAN program developed by the research team has broken this limit, and realized the Raman spectrum quantum simulation of COVID-19 spike protein containing more than 100 million atoms in aqueous solution for the first time. This breakthrough is attributed to the team's multiple innovations in algorithm design and engineering technology. In traditional density functional theory (DFT) and density functional perturbation theory (DFPT) calculations, the computational complexity increases exponentially with the size of the system, which limits the calculations to small systems. In response to this issue, the team has developed a new method that deeply integrates all electron full potential density functional perturbation theory with quantum partitioning algorithm. By decomposing complex biomolecules into multiple subsystems, the computational complexity is significantly reduced. At the same time, the team has developed a multi-level scheduling technology that is sensitive to block size to address the load balancing challenge of massive block computing, improving the parallel scalability of massive block computing; Elastic task offloading technology is proposed to address the heterogeneous acceleration challenges of small-scale operations; By flexibly aggregating small-scale operations, the hardware utilization of heterogeneous accelerators has been significantly improved. In addition, the QF-RAMAN program adopts the OpenCL universal heterogeneous parallel computing framework, which can run across platforms on supercomputers with different hardware architectures (CPU, GPU, SW, etc.) using the OpenCL compilation toolchain (oneAPI, rocm, swcl, etc.). On the latest generation of Shenwei supercomputer, the program utilizes 96000 computing nodes (over 37 million computing cores) to achieve dual precision peak performance of 399.9 PFLOP/s; On the Eastern supercomputer, using 6000 nodes (24000 GPUs), it also demonstrated excellent performance of 85 PFLOP/s. The strong and weak scalability performance of both supercomputers are close to the ideal value, fully demonstrating the efficiency and scalability of this method. On this basis, the team proposed a new algorithm for solving Raman spectra using matrix equations applicable to billion level atomic systems, which avoids direct diagonalization and provides a new solution for high-precision Raman spectroscopy calculations, effectively solving key technical problems in large-scale quantum mechanical Raman simulations.

　　This study not only demonstrates China's leading position in high-performance computing and computational chemistry, but also extends quantum mechanics simulation to unprecedented computational scales, opening up new avenues for understanding complex biological systems and exploring more application scenarios for quantum mechanics simulation.