Quantitative analysis of EXAFS data sets using deep reinforcement learning

Extended X-ray absorption fine structure (EXAFS) serves as a unique tool for accurately characterizing the local structural properties surrounding specific atoms. However, the quantitative analysis of EXAFS data demands significant effort. Artificial intelligence (AI) techniques, including deep reinforcement learning (RL) methods, present a promising avenue for the rapid and precise analysis of EXAFS data sets. Unlike other AI approaches, a deep RL method utilizing reward values does not necessitate a large volume of pre-prepared data sets for training the neural networks of the AI system. We explored the application of a deep RL method for the quantitative analysis of EXAFS data sets, utilizing the reciprocal of the R-factor of a fit as the reward metric. The deep RL method effectively determined the local structural properties of PtO_x and Zn-O complexes by fitting a series of EXAFS data sets to theoretical EXAFS calculations without imposing specific constraints. Looking ahead, AI has the potential to independently analyze any EXAFS data, although there are still challenges to overcome.