Deep learning-based neural network potential for investigating the synergistic effect of H and He in BCC-Fe

Reduced activated ferrite/martensite (RAFM) steel is a prospective structural material for fusion reactors. The interaction of deuterium/tritium fusion neutrons with structural materials leads to transmutation, resulting in the production of hydrogen (H) and helium (He) atoms. These atoms lead to a synergistic effect that severely damages material integrity. Molecular dynamics (MD) simulations, which depend on precise interatomic potentials, are crucial for understanding this phenomenon. In our research, a Deep Potential (DP) method-based machine-learning (ML) potential has been developed for the Fe-H-He ternary system, which is referred to as DP-FeHHe. For body-centered cubic (BCC) Fe, the potential demonstrates quantitative in predicting its basic properties, the behavior of solute H and He atoms, the formation energy of H₂ molecules, and the interactions of small He/H clusters with each other and with vacancy defects. For larger He/H bubbles in BCC Fe, the potential can qualitatively describe their general behavior and trends. Using the DP-FeHHe potential, our simulations show that the critical size of He cluster for emitting an Fe atom decreases with increasing temperature, which is consistent with other models. Furthermore, at 300 K, we observed H atoms gathering on the surface of a 0.8 nm diameter He-vacancy cluster during a 4 ns simulation, forming a core/shell structure. This observation is consistent with experimental phenomena. This demonstrates the reliability of our potential for exploring the synergistic effects of H and He in Fe.