Modelling Point Defects at Finite Temperatures with Machine Learning Force Fields

Point defects control the properties of most functional materials. Their identification is addressed by combining experimental measurements with theoretical models. Most ab-initio studies are limited to a 0 K description due to the high computational cost. In this study, we move beyond this idealised model and explore the effects of finite temperatures on defect formation, using Te_i in CdTe as an exemplar. By including the main entropic contributions, like configurational (structural, spin) and vibrational, we find that thermal effects increase the predicted concentrations by three orders of magnitude. Further, to study the defect dynamics over longer time and length scales we use a machine learning force field (MLFF). These simulations reveal that Te_i rapidly exchanges between configurations and sites already at room temperature, where the accessible low-energy configurations act as stepping stones that promote diffusion. Finally, we assess the interpretability of our MLFF, observing strong correlation between the atomic energies and the defect-induced distortion. Overall, our study underscores the importance of finite-temperature effects and the potential of MLFFs to accurately model complex defect processes.<br/><br/>I. Mosquera-Lois, S. R. Kavanagh, J. Klarbring, K. Tolborg & A. Walsh, <i>Chem. Soc. Rev.</i>, <b>52</b>, 5812-5826 (2023)