C++Programming for ReaxFF-nn Machine Learning Potential Model Implementation in LAMMPS
This paper introduces the C++programming of ReaxFF-nn(neural network)machine learning potential model in LAMMPS,a large-scale molecular dynamics parallel computational simulation software developed by Sandia National Laboratories in the U.S.A.It has the advantages of high computational efficiency,many potential fields,and open source code with good parallel scalability.ReaxFF is based on bond-level reaction force fields.In this work,the bond level and bond energy equations are redesigned using neural network on the basis of ReaxFF module in LAMMPS to form the machine learning potential ReaxFF-nn.This paper explains in detail the basic principle of ReaxFF-nn and the programming flow of its implementation in LAMMPS,including reading input data,initializing the parameters of the neural network,calculating the inter-atomic bond energies as well as functions of bond levels,and other steps.The accuracy and reliability of the calculations via ReaxFF-nn are verified through example simulations.
molecular dynamicsReaxFF-nnLAMMPSmachine learningreaction force field
万方数据
