Publication date: 21st July 2025
The combination of machine learning (ML) with density functional theory (DFT) accelerates material simulations, expanding both spatial and temporal scales. However, current ML methods struggle to address polaron trapping. Polarons are quasi-particles arising from electron-phonon coupling in a wide range range of materials and shape the properties of the hosting systems. Therefore, understanding polaron effects is key for technonlogical applications. We present a novel machine learning force field (MLFF) approach that incorporates polaron trapping descriptors, enabling large-scale studies of polaronic materials.
Using TiO$_2$(110) as a case study, we reveal how dopants and atomic vacancies affect polaron configurations and drive catalytic CO adsorption. Additionally, our method captures the dynamic evolution of polarons with unprecedented statistical robustness.
This work advances fundamental understanding of defect-polaron interactions while offering a fully automated and efficient computational suite for the study of polaronic materials, facilitating characterization and design of metal oxide catalysts.