Creation of a Two-Step Prediction Model of 4f-5d Transition Energies of Ce3+ in Garnet-Type Oxides Based on the Combination of First-Principles Calculations and Machine Learning

Abstract

Listen

Translate article

This study presents a novel approach which combines first-principles calculations with machine learning techniques to predict the 4f-5d transition energy of Ce3+ ions in garnet-type oxides. This is important in various technological applications, e.g., light-emitting materials, and solid-state lighting devices. However, it is difficult and time-consuming to accurately determine this energy using conventional methods. A linear regression model for the 4f-5d transition energy of Ce3+ in garnet-type oxides was created by machine learning, using electronic and structural parameters as attributes. The electronic parameters were obtained through first-principles calculations. This model is highly effective in estimating transition energy values for various Ce3+-doped garnet-type oxides. The machine learning model improved the accuracy of transition energy calculations. Furthermore, we have developed predictive models that estimate first-principles calculation results based on structural data. Systematic first-principles calculations of fictitious garnet-type oxides with gradually changed structural parameters were performed and used as the training data. The predictive models of the electronic parameters such as the bond order, the crystal field splitting (εcfs) and the net charge were used to create a two-step predictive model based solely on structural parameters.