Modeling Complex Multicomponent Materials with First-Principles Based Statistical Mechanics and Machine Learning

Abstract

Recent progress in the engineering of multicomponent, solid-state compounds for optoelectronic applications has entailed an ever expanding range of material chemistries and a rapid increase in material complexity. For example, within the classes of chalcogenide and halide-perovskite semiconductors, fundamental material properties can be effectively tuned by alloying various isovalent chemical species and by the controlled incorporation of dopants. In characterizing these materials at an atomistic level, one has to contend not only with the presence of a multitude of point defects, but also with the potential formation of ordering and instabilities against secondary phases. This poses a fundamental challenge to first-principles modeling that is only exacerbated by an exponentially large configuration space. This work is primarily concerned with the development and application of methods, rooted in statistical mechanics and machine learning, for modeling these complex, multicomponent semiconductors. Much of this manuscript focuses on modeling defects and configurational disorder in specific chalcogenides and halide-perovskites of particular technological relevance, employing well-established multiscale methods such as density functional theory, point defect thermodynamics, statistical learning, cluster expansions and Monte Carlo simulation. The latter portion of this work concerns more recent developments in the fields of deep learning and tensor networks, adapted and applied to material structure-property prediction.