Development of a multi-element neural network modified lattice inversion potential and application to the Ta-He system

Under extended radiation exposure and elevated temperatures, helium (He) accumulation can compromise the integrity of Tantalum (Ta), a material showing substantial promise for nuclear fusion reactor applications. An imperative step towards understanding and enhancing the performance of Ta-based materials lies in the development of an accurate potential for Ta-He interactions. In this study, we introduce a neural network modified lattice inversion potential (NN-LIP) specifically designed for nuanced, element-specific Ta-He interactions. The conventional atomic density descriptors have been augmented to encompass sub-atomic characteristics, ensuring an accurate representation of varied local chemical environments across the energy spectrum. The lattice inversion potential for individual atoms introduces an atomic cross-potential, bolstering the transferability and robustness of NN-LIP, thereby amplifying its extrapolation capacity. Empirical calculations centered on pivotal properties of the Ta-He system affirm the precision of the multi-element NN-LIP potential. Our comprehensive dataset, spanning 19162 samples, covers a gamut from Ta bulk properties to point defects, He vacancies, and migration traits, achieving a validation set accuracy of 5.14 meV/atom. Notably, the migration barrier prediction accuracy displayed marked improvement over prior studies. The formulated multi-element NN-LIP offers a detailed examination of Ta-He interactions and holds potential for modeling analogous metalhelium interactions in nuclear substrates.