Chinese team wins Gordon Bell Prize for nitrogen-fixing enzyme simulation


The Association for Computing Machinery recognized what it called the first exascale classical solution to a molecular simulation problem. The work models part of an enzyme involved in converting nitrogen into ammonia.
The Association for Computing Machinery (ACM) awarded its 2026 Gordon Bell Prize to a ten-member team from Chinese institutions for what it called the first exascale classical solution to the FeMoco complete-active-space problem. ACM presented the prize at the SC26 computing conference in Chicago on 20 November 2026.
The result follows earlier classical work on the same kind of model: in 2026, Huanchen Zhai and colleagues estimated its ground-state energy to chemical accuracy using coupled-cluster and density-matrix renormalization-group calculations with extrapolation. The prize-winning team reports a solution at exascale—a rate of at least a quintillion floating-point operations per second.
The team studied a model of the electronic ground state, or lowest-energy state, of FeMoco, the active site of nitrogenase, rather than the full process by which the enzyme enables organisms to convert atmospheric nitrogen into ammonia.
The model’s complete active space covers 113 electrons in 76 orbitals, regions associated with possible electron states. The method considers configurations of the selected electrons and orbitals while treating other orbitals separately. The team said its calculation reached a consensus ground-state result for that active space.
The researchers tuned their code for two supercomputers, LineShine and China’s New-generation Intelligent Supercomputer. On 20,480 LineShine nodes, they reported sustained rates, achieved during computation rather than at the machine’s theoretical maximum, of 1.456 exaflops in double precision and 5.626 in mixed precision. The double-precision rate was 58.9% of the machine’s peak.

Double precision uses a 64-bit number format; mixed precision uses a different precision category, so the two rates are not directly equivalent.
In a separate calculation, the team said it reached the ground-state consensus in a measured end-to-end time of 105 minutes on 1,280 LineShine nodes. The paper compared that measured result with a published 8.6-hour single-run projection, rather than a measured run, for a fault-tolerant quantum computer on the same active space.
ACM named the winners as Honghui Shang, Zhiya Xiao, Zhuozhao Xia, Hao Wu, Huan Ma, Yongqing Chen, Bin Li, Zhenyu Li, Yutong Lu and Jinlong Yang. The conference proceedings spell Xiao’s given name Zhiyu; both list Xiao at the Chinese Academy of Sciences.
The award announcement lists Zhenyu Li at Sun Yat-sen University, while the proceedings list him at the University of Science and Technology of China.
The nuclear technology company also plans a $280 million Tennessee campus for manufacturing and fueling, with construction already under way and first commercial deliveries targeted for 2028.
2 min

The classification society believes its framework is the first of its kind suitable for every type of ship and offshore unit. Developers can use it to assess designs, but individual projects still need separate authorization.
3 min

The longer process applies to a specific ChatGPT setup. Apple says its new controls help users understand the risks of granting broad access, while some companies and users see an effort to slow competitors.
2 min