Understanding Imperfections and Instabilities in Crystals via Physics-Based and Data-Driven Models

Citation

Teh, Ying Shi (2021) Understanding Imperfections and Instabilities in Crystals via Physics-Based and Data-Driven Models. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/kd3n-eq78. https://resolver.caltech.edu/CaltechTHESIS:04202021-184720643

Abstract

In crystals, atoms are arranged in a periodic manner in space. However in reality, imperfections and instabilities exist and this repeated arrangement is never perfect. The coupling between crystal defects, lattice instabilities, other defects like domain walls and domain patterns, and material properties generates interesting phenomena that can be leveraged on for future materials design. Nevertheless, the coupling of different scales and processes also makes the modeling and understanding of these materials an open challenge. This thesis examines these various aspects of crystalline solids through the development of both physics-based and data-driven computational models at the appropriate length scales.

Above-bandgap photovoltaic (PV) effect has been observed experimentally in multi-domain ferroelectric perovskites, but the underlying working mechanisms are not well understood. The first part of the thesis presents a device model to study the role of ferroelectric domain walls in the observed PV effect. The model accounts for the intricate interplay between ferroelectric polarization, space charges, photo-generation, and electronic transport. When applied to bismuth ferrite, results show a significant electric potential step across both 71° and 109° domain walls, which in turn contributes to the PV effect. The domain-wall-driven PV effect is further shown to be additive in nature, allowing for the possibility of generating the above-bandgap voltage.

In the second part, we present a lattice model incorporating random fields and long-range interactions where a frustrated state emerges at a specific composition, but is suppressed elsewhere. The model is motivated by perovskite solid solutions, and explains the phase diagram in such materials including the morphotropic phase boundary (MPB) that plays a critical role in applications for its enhanced dielectric, piezoelectric, and optical properties. Further, the model also suggests the possibility of entirely new phenomena by exploiting MPBs.

The final part of the thesis focuses on constructing data-driven models from first principles calculations, particularly density functional theory (DFT) for studying crystalline materials. Specifically we propose an approach that exploits machine learning to approximate electronic fields in crystalline solids subjected to deformation. When demonstrated on magnesium---a promising light weight structural material---our model predicts the energy and electronic fields to the level of chemical accuracy, and it even captures lattice instabilities. This DFT-based machine learning approach can be very useful in methods that require repeated DFT calculations of unit cell subjected to strain, especially multi-resolution studies of crystal defects and strain engineering that is emerging as a widely used method for tuning material properties.

Thesis Files