Machine Learning Nonadiabatic Molecular Dynamics Reveals Key Influence of Topology of Oxygen Vacancies on Carrier Trapping and Recombination in SrTiO3

Common defects in numerous systems, oxygen-vacancies (VO) shape trap states and govern charge carrier recombination, yet the role of their spatial topology remains poorly understood. We demonstrate, using machine-learning nonadiabatic molecular dynamics, a strong influence of VO dimer connectivity on charge carrier trapping and recombination in a widely used SrTiO3 oxide. Focusing on discrete (2 VO-D), bent (2 VO-B), and linear (2 VO-L) vacancy dimers, we show that strong interactions between VO vacancy complexes at high VO concentrations perturb the defect-state wave functions. All three configurations form shallow traps near the conduction-band minimum (CBM) and exhibit similar nonradiative recombination with lifetimes of hundreds to thousands of nanoseconds. At low VO concentrations, the intrinsic topology becomes decisive. The 2 VO-D and 2 VO-B configurations retain spatially distributed and asymmetric defect states, and remain shallow, whereas the symmetric 2 VO-L configuration stabilizes a localized in-gap state through enhanced Ti-centered orbital hybridization. This deep trap opens an efficient CBM → trap → VBM (valence band minimum) pathway and markedly accelerates nonradiative recombination. The results establish vacancy topology and defect-wave function symmetry as key microscopic determinants of trap depth and carrier lifetime, while highlighting the importance of sufficiently large models for realistic defect-dynamics simulations.