Highly accurate and efficient potential for bcc iron assisted by artificial neural networks

Atomic forces and energies, calculated by interatomic potential, are fundamental components of molecular dynamics (MD) and Monte Carlo (MC) simulations. Compared with traditional potentials, machine-learning (ML) potentials trained by using extensive density-functional theory databases exhibit high accuracy in predicting physical and chemical properties of materials, but their transferability often faces constraints. To address this limitation, physically informed neural network (PINN) potentials have been developed. These models synergistically combine the strengths of ML with physics-based bond-order interatomic potentials, aiming for both improved accuracy and broader applicability. However, a major limitation remains: the low performance of PINN potentials, hindering large-scale simulations. This work introduces a potential framework by incorporating an artificial neural network (ANN) into typical potential functions, which not only improves the transferability compared with the ANN potential, but also significantly improves the performance of ML potentials. The developed ANN assistant potential for body-centered cubic (bcc) iron demonstrates exceptional accuracy in property predictions while boasting remarkable computational efficiency. Its performance utilizing a single graphics processing unit (GPU) card overcomes both 12-message passing interface central processing unit -only ML potential and GPU-accelerated ML potential by achieving speedups of 201\ifmmode\times\else\texttimes\fi{} and 26\ifmmode\times\else\texttimes\fi{}, respectively. The proposed approach has a potential to provide a powerful way to develop high accurate and efficient potentials even in the other systems.