An improved reactive force field parameter optimization framework based on simulated annealing and particle swarm optimization algorithms

Abstract

Atomic-scale simulations are important tools for microscopic phenomena study and material design, especially the cost-effective and large-scale reactive force field (ReaxFF). However, the poor transferability and tedious training process of ReaxFF parameters constrain its accuracy and application, urgently requiring more efficient automatic optimization methods. In this study, we propose a multi-objective optimization method that combines simulated annealing algorithm (SA) and particle swarm optimization algorithm (PSO) to optimize the ReaxFF parameters. Moreover, we innovatively introduce a concentrated attention mechanism (CAM) to improve the accuracy of parameter optimization. Finally, this study selects the H/S system as the testing target to evaluate the accuracy and efficiency of the above algorithm. It is found that our algorithm is faster and more accurate than traditional metaheuristic methods. Our automated optimization scheme efficiently optimizes ReaxFF parameters, providing crucial support for atomic-scale simulations.