Xie Xin-Tian, Yang Zheng-Xin, Chen Dongxiao, Shi Yun-Fei, Kang Pei-Lin, Ma Sicong, Li Ye-Fei, Shang Cheng, Liu Zhi-Pan
Collaborative Innovation Center of Chemistry for Energy Material, Shanghai Key Laboratory of Molecular Catalysis and Innovative Materials, Key Laboratory of Computational Physical Science, Department of Chemistry, Fudan University, Shanghai 200433, China.
State Key Laboratory of Metal Organic Chemistry, Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, Shanghai 200032, China.
Precis Chem. 2024 Sep 14;2(12):612-627. doi: 10.1021/prechem.4c00060. eCollection 2024 Dec 23.
Atomic simulations aim to understand and predict complex physical phenomena, the success of which relies largely on the accuracy of the potential energy surface description and the efficiency to capture important rare events. LASP software (large-scale atomic simulation with a Neural Network Potential), released in 2018, incorporates the key ingredients to fulfill the ultimate goal of atomic simulations by combining advanced neural network potentials with efficient global optimization methods. This review introduces the recent development of the software along two main streams, namely, higher intelligence and more automation, to solve complex material and reaction problems. The latest version of LASP (LASP 3.7) features the global many-body function corrected neural network (G-MBNN) to improve the PES accuracy with low cost, which achieves a linear scaling efficiency for large-scale atomic simulations. The key functionalities of LASP are updated to incorporate (i) the ASOP and ML-interface methods for finding complex surface and interface structures under grand canonic conditions; (ii) the ML-TS and MMLPS methods to identify the lowest energy reaction pathway. With these powerful functionalities, LASP now serves as an intelligent data generator to create computational databases for end users. We exemplify the recent LASP database construction in zeolite and the metal-ligand properties for a new catalyst design.
原子模拟旨在理解和预测复杂的物理现象，其成功很大程度上依赖于势能面描述的准确性以及捕捉重要罕见事件的效率。2018年发布的LASP软件（基于神经网络势的大规模原子模拟）通过将先进的神经网络势与高效的全局优化方法相结合，融入了实现原子模拟最终目标的关键要素。本综述沿着两个主要方向介绍了该软件的最新进展，即更高的智能性和更多的自动化，以解决复杂的材料和反应问题。LASP的最新版本（LASP 3.7）采用了全局多体函数校正神经网络（G-MBNN），以低成本提高势能面的准确性，实现了大规模原子模拟的线性缩放效率。LASP的关键功能进行了更新，纳入了：（i）用于在巨正则条件下寻找复杂表面和界面结构的ASOP和ML-接口方法；（ii）用于识别最低能量反应路径的ML-TS和MMLPS方法。凭借这些强大的功能，LASP现在作为一个智能数据生成器，为终端用户创建计算数据库。我们举例说明了最近在沸石中构建LASP数据库以及用于新催化剂设计的金属-配体性质。
