Data-Driven Design and Exploration: Advancing Materials Science Through Computational Methodologies and Machine Learning
This thesis investigates the cutting-edge intersection of Computational Materials Science and data-driven methodologies, focusing on the exploration, understanding, and design of materials. Leveraging tools such as Density Functional Theory, Machine Learning, and Bayesian optimization, the research delves into various domains of materials science. In the pursuit of spintronic devices, a novel database of ab initio calculated spin splitting in 2D materials was established, marking the first of its kind in Brazil. The research further proposed an innovative workflow integrating inverse design with Bayesian inference optimization for materials design. A significant contribution was made to the understanding of large Rashba spin splitting in crystalline solids, emphasizing anti-crossing as a design principle. The study also explores the transformation from traditional trial-and-error methods to intelligent, data-based strategies in designing functional materials, particularly magnetic 2D compounds. Utilizing databases, ab-initio calculations, and machine learning algorithms, the research provided key insights into magnetic order in those materials, expanding possibilities for stable and easily synthesized compounds. In addition, the research utilized data from OQMD to construct a graph-based recommender system, leading to the proposal of novel compounds and illustrating a novel method for materials discovery. Together, these advancements contribute significantly to the scientific community, offering new insights and methodologies in materials science. The thesis stands as a guide to future research in this dynamic field, aiming to inspire further exploration and innovation, and presenting an expanded framework for our understanding and mastery of the materials that constitute the technologies we have around us.