Materials exploration by optimization of chemical composition using first-principles calculation and Bayesian optimization

Abstract

We are developing methods for exploring magnet compounds using machine learning techniques and first-principles calculation. In this presentation, we discuss a framework of optimization we have recently proposed based on a machine learning technique called Bayesian optimization. We will show that our framework can efficiently optimize magnetization, Curie temperature, and formation energy-which are key performance indicators of magnet compounds-with respect to the choice of additive elements and their amount.