Prediction And Investigation of Transition Metal Oxides for Hydrogen Storage Application via Machine Learning-Assisted Density Functional Theory

This study presents a Density Functional Theory (DFT) investigation of transition metal oxides for hydrogen storage, employing machine learning as a supplementary screening tool. Given the scarcity of TMO data, we leverage Unsupervised Domain Adaptation to transfer knowledge from metal hydrides to metal oxides. The Nearest Neighbor Weighting with eXtreme Gradient Boost (NNW-XGB) model demonstrated superior screening performance, ranking promising candidates and predicting hydrogen storage capacities of 3.23 Hwt% for spinel Ca₂TiO₄ and 3.14 Hwt% for perovskite CaVO₃. DFT validation using the PBE0 + Grimme's D3 confirmed that both materials maintain structural and electronic stability up to their saturation limits. The comparison shows good agreement for Ca₂TiO₄ (3.03 Hwt% vs. 3.23 Hwt%) but an overestimation for CaVO₃ (2.11 Hwt% vs. 3.14 Hwt%), attributed to lattice strain effects not captured by composition-based descriptors. Detailed electronic structure analysis via projected density of states (PDOS) and d-band center calculations revealed that hydrogen adsorption is driven by charge transfer from hydrogen to the B-site transition metal (Ti, V), with the progressive filling of d-states leading to electronic saturation at higher coverages. Compositional trend analysis identified promising combinations of A-site (Mg, Be, Ca, Al) and B-site (Sc, Ti, V, Fe) elements.