An improved meta-heuristic algorithm for developing high-quality ReaxFF force fields of Fe/Ni transition metals and alloys

Abstract

A reliable and transferable reactive force field (ReaxFF) is necessary for ReaxFF molecular dynamic simulations to study physical and chemical interactions of interest. The determination of ReaxFF force fields is a particularly challenging global optimization problem with multiple local minima. Here, an improved meta-heuristic algorithm (IMHA) based on particle swarm optimization (PSO) is proposed to accelerate the convergence of the optimization solution and improve the quality of ReaxFF parameterization. Firstly, a Latin hypercube design (LHD) algorithm is used to generate a uniform distribution of initial swarm, improving the exploration of whole parameter space. Secondly, the strategies for generating trial individuals in backtracking search optimization are modified and applied into PSO to enhance the search capacities and evade from local optimum. Thirdly, the sequential one parameter parabolic extrapolation method is utilized to obtain a rapid convergence in local region by optimizing global best particle. Consequently, the IMHA algorithm is demonstrated its capabilities by optimizing ReaxFF force field parameters of Fe/Ni transition metals and alloys. MD simulations show that the developed force fields accurately reproduce the fundamental properties of bcc Fe and fcc Ni transition metals and their alloys compared to experimental and QM values.