Hot Carrier Relaxation in Halide Double Perovskites: Insights from <i>Ab Initio</i> Models and Structure-Informed Machine Learning

The hot carrier (HC) cooling at an ultrafast timescale represents a critical limitation in achieving high-efficiency photon-to-electron conversion. The comprehensive materials screening seeks to identify candidates that inherently suppress hot carrier relaxation, thereby extending carrier lifetimes. Here, we combine time-dependent density functional theory and nonadiabatic molecular dynamics to investigate the hot electron (HE) dynamics in experimentally synthesized Cs2AgInX6 (X = Br and Cl). These halide double perovskites exhibit significantly extended HE lifetimes compared to conventional metal halide perovskites, primarily due to the presence of discrete electronic states at the conduction band edge and weakened. The well-trained machine learning models, including artificial neural networks (ANN) and mutual information (MI), efficiently capture the complex relationships among composition, structural dynamics, and excited-state properties while drastically reducing computational expense. The ANN models that are trained using only 10% of data enable accurate interpolation of dynamic electronic properties such as energy gaps near the conduction band edge and nonadiabatic coupling. The high accuracies of predicted properties ensure the ability of these ANN models to capture transient changes in electronic structure over extended timescales. The ANN-generated interpolated data enhance the reliability of MI analyses that uncover subtle, time-resolved correlations between structural dynamics and HE relaxation. These insights pin down the strategic design approaches to realize stable materials for HE-based optoelectronic applications.