Unveiling hydrogen chemical states in supersaturated amorphous alumina via machine learning-driven atomistic modeling

Abstract Advancing hydrogen-based technologies requires detailed characterization of hydrogen chemical states in amorphous materials. As experimental probing of hydrogen is challenging, interpretation in amorphous systems demands accurate structural models. Guided by experiments on atomic layer deposited alumina, a fast atomistic simulation technique is introduced using an ab initio-based machine learning interatomic potential to generate amorphous structures with realistic hydrogen contents. As such, the annealing of highly defective crystalline hydroxide structures at atomic layer deposition temperatures reproduces experimental density and structure, enabling accurate prediction of Al Auger parameter chemical shifts. Our analysis shows that higher hydrogen content favors OH ligands, whereas lower hydrogen content leads to diverse chemical states and hydrogen bonding, consistent with charge density and partial Bader charge calculations. Our approach offers a robust route to link hydrogen content with experimentally accessible chemical shifts, aiding the design of next-generation hydrogen-related materials.