Ab-initio and machine-learning accelerated atomistic simulations for TiFe hydrogen storage applications

In this thesis, atomistic modeling and simulation methods are thoroughly employed to understand and explore the mechanism of oxidation and subsequent hydrogenation of the TiFe intermetallic compound. Parallel to the material investigations, the computational methodology has also advanced with the incorporation of machine-learning techniques. TiFe is a commercially utilized stationary hydrogen storage material with attractive properties for practical applications. The formation of oxide-films on the TiFe surface is a major concern limiting its wide application. The first chapter of this work constructs and characterizes the oxygen-exposed TiFe surface with the use of global-optimization techniques and ab-initio methods. The energetics and dynamics of early-oxidation of the material surface is studied at different temperatures and the activation barriers for hydrogen penetration through the oxide layers are evaluated with a focus on the activation mechanisms. Structure fingerprint tools employed are able to identify metal oxide structural motifs consistent with experimental findings. This work also produces valuable mechanistic insights for the activation of TiFe. Progressively, the influence of transition metal dopants, V, Cr, Mn and Ni, substituting Fe in TiFe is investigated with a focus on their relative preference to oxidize and their involvement in the surface oxidation. Their suitability to promote hydrogenation is evaluated and characterized by assessing the change in surface chemistry and through a comparison with the pristine TiFe surface. Accurate atomistic simulations are burdened by heavy computational cost and limited to small systems. Hence, in the last chapter of this thesis, an extensive machine-learning interatomic model with ab-initio accuracy is developed for Ti-Fe-H system utilizing the first-principles data generated throughout this work. This deep neural network model provides upto 1000 times faster computation and scales linearly with the size of the system bridging the gap between ab-initio and large-scale atomistic simulations.