A general deep neural network potential model for transition metal carbides

Abstract Transition metal carbides (TMCs) are well‐known for their remarkable properties, such as high melting points, excellent oxidation and corrosion resistance, and outstanding mechanical properties. These properties make them ideal candidates for applications in extreme environments. Consequently, there is a pressing need for reliable and efficient material design tools to expedite the development of new TMCs. To address this demand, we have developed a domain‐specific, medium‐scale interatomic potential model for TMCs that encompasses a wide range of elements, including Ti, Zr, Hf, V, Nb, Ta, Cr, Mo, W, C, and N, and covers three typical crystal structures as well as melting structures and surface structures. Our model demonstrates predictive accuracy on par with density functional theory‐based methods, with errors in energy and force being 9.2 meV/atom and 363 meV/Å, respectively. Moreover, the model demonstrates high accuracy in predicting various material properties, such as lattice parameters, elastic constants, equations of state, and melting points. It also accurately predicts grain boundary segregations and stacking fault structures near the hetero‐phase interface. By offering a reliable and efficient tool for material design, our model will play an instrumental role in the development of TMCs, ultimately benefiting various industries that rely on TMCs.