DFT dataset for Fe-TiC-H machine learning interatomic potentials

The trapping of hydrogen by TiC and other transition-metal carbides is seen as a possible solution to hydrogen embrittlement of steels. The trapping capability of carbides depends on their structure and chemistry, as well as the morphology of the carbide/matrix interface. The interaction of hydrogen with such nano-scale features is extremely challenging to investigate experimentally. First-principles methods, such as DFT, have partially addressed this limitation by calculating accurate energetics for hydrogen in bulk carbides. However, the extended nature of the carbide/matrix interface makes its treatment with DFT computationally prohibitive. We have developed a machine learning interatomic potential within the framework of Moment Tensor Potentials, that can be used to predict the energetics and kinetics of hydrogen in bcc Fe, rock salt TiC, and the TiC/Fe interface. The main contribution of this potential is to probe the segregation and migration of hydrogen at the semi-coherent and incoherent TiC/Fe interfaces, including features such as coherency strains, misfit dislocations, and strain-induced vacancies. This dataset contains the DFT training dataset and the trained potential.