Deep potential-driven deciphering of microstructure-corrosion relationships in MgNi alloys

Abstract The design of materials for oil and gas extraction is critical to the long-term safe operation of downhole equipment and the exploitation of deep-sea and unconventional resources. MgNi alloys exhibit high strength and excellent corrosion resistance, but their unsatisfactory mechanical performance, such as low stiffness and inherent brittleness, restricts practical applications and processing. In this work, a deep potential (DP) model for MgNi alloys was developed within the AI for Science framework to investigate the effects of composition and point defects on crystallographic, mechanical, and corrosion behaviors. The DP model achieves high accuracy, with root-mean-square errors values of 1.48 × 10 −3 eV/atom for energy and 3.47 × 10 −3 eVÅ –1 for atomic forces. Predicted lattice constants and mechanical properties agree well with density functional theory and experimental results, with deviations within 0.5%. Key energy parameters are reproduced with relative errors below 1%, demonstrating reliable simulation capability for corrosion dissolution kinetics. Overall, the proposed DP model provides an efficient and accurate tool for performance optimization and reliability design of MgNi alloys, and offers a robust basis for multiscale modeling and inverse design of high-performance alloys.