The objective of this collaborative research is to apply computational thinking to materials science with the goals of revealing hidden rules about materials structure and properties and providing efficient computational and statistical tools for modeling large systems with interacting elements. The approach combines materials science with statistical learning. The research is driven by two key problems in materials development, crystal structure prediction and the inverse problem in materials science whereby one postulates the desired properties and finds the composition and arrangement of atoms that result in those properties. The investigators seek to design principled Bayesian models for relational data, coupled with efficient inference methods.
With respect to intellectual merit, the integrative approach is a significant departure from current methods for materials research. Ab initio computation has begun to show promise for materials development, but its integration with statistical learning holds the promise of leading to novel approaches that can utilize massive amounts of materials data. Further, extracting knowledge from massive relational data presents opportunities for machine learning research. The research addresses common challenges in many disciplines and provides new mathematical frameworks and computational tools.
With respect to broader impacts, the research has the potential to enhance materials research and, ultimately, lead to the development of better materials. The application focus on materials for energy is timely and important. The investigators plan to recruit women and other students from underrepresented groups into their research teams. Results will be disseminated through education and a cyber-based platform, exposing computer science students to engineering applications.
