Revisão teórica da previsão de band gap em perovskitas por meio de modelos de Machine Learning e Teoria do Funcional da Densidade (DFT)

Abstract

This work presents a theoretical review of band gap prediction in perovskites, focusing on the integration between traditional first-principles methods, represented by Density Functional Theory (DFT), and Machine Learning (ML) models applied to materials science. Initially, the theoretical foundations of solid materials are postulated, such as crystalline structures and factors that influence the band gap. This discussion continues with a description of the particularities of perovskites, their structure, and optoelectronic behavior. Next, the formulation of DFT is explored, along with its practical limitations associated with exchange and correlation functional approximations and the high computational cost in extensive systems. Based on this, the relevance of Machine Learning models as an alternative to mitigate these limitations is presented, with a comparative analysis of different algorithms used in perovskite band gap prediction, including Random Forest, Support Vector Regression, and Boosting methods. The results reported in the literature indicate that these models achieve competitive performance metrics when trained on suitable datasets, reinforcing the potential of the hybrid approach between DFT and ML. Finally, it is discussed the importance of the continuous expansion of datasets for better interpretability of the models, the careful selection of descriptors, and the investigation of compositional engineering strategies for consolidating the advantages of perovskites as promising materials for emerging optoelectronic technologies.