Machine Learning-Directed Predictive Models: Deciphering Complex Energy Transfer in Mn-Doped CsPb(Cl1–yBry)3 Perovskite Nanocrystals

Abstract

Lead halide perovskite nanocrystals with inclusion of a transition-metal dopant of Mn2+ offer a substantial degree of freedom to modulate the optoelectronic and magnetic properties owing to the introduced dopant in the host lattices. However, complexity as a result of the excited interactions between the exciton and dopant, involving dynamics of exciton recombination, competing forward and backward energy transfer (and vice versa), and Mn recombination, makes it difficult to understand and predict the Mn sensitization. Here, we have created machine learning-directed models using different nonlinear algorithms with initial 86 samples to decipher the complex energy transfer by navigating the reaction design space of various concentrations of Mn along with different halide compositions (band gap) in Mn-doped CsPb(Cl1–yBry)3 nanocrystals. K-nearest neighbor-based predictive models coupled with time-correlated single photon counting measurements allow for fully elucidating the complex and competing energy transfer kinetics occurring in two different Mn concentration regimes. Importantly, forward exciton-to-Mn energy transfer is more governed by the Mn concentration, while the backward Mn-to-exciton energy transfer is strongly dependent on the energy gap difference between the exciton and Mn energy state. This machine learning-guided approach and modeling can not only provide an efficient means for navigating the vast reaction design space but also provide significant insight into understanding and elucidating the complex physical phenomena throughout analyzing and predicting the dataset trend.