Deep learning potential-driven study of multiscale structural and thermodynamic behaviors in PtTi alloys

Abstract

PtTi alloys exhibit great potential for applications in key fields such as aerospace, chemical engineering, and energy due to their excellent high-temperature performance, catalytic activity, and corrosion resistance. However, fundamental research on PtTi alloys remains limited, particularly regarding their structural characteristics, performance optimization, and critical issues in practical applications, which have yet to be fully addressed. Therefore, this study proposes a universal machine learning potential for PtTi alloys by integrating neural networks and active learning methods. A high-quality training dataset was established through first-principles calculations. The machine learning potential developed in this work enables efficient and reasonably accurate predictions of the crystal structure, thermodynamic properties, stacking fault energy, and tensile behavior of PtTi alloys, showing good performance in terms of both accuracy and computational efficiency. Compared with traditional first-principles calculations, our approach significantly enhances computational speed while maintaining accuracy, enabling large-scale molecular dynamics simulations. This provides a precise and efficient tool for exploring the multiscale behavior of alloys. Furthermore, this study offers a solid theoretical foundation and technical support for optimizing the high-temperature performance, structural design, and future engineering applications of PtTi alloys. By employing this innovative approach, we pave new avenues for enhancing the performance and expanding the applications of PtTi alloys, highlighting its significant scientific value and practical potential.