Towards Virtual Autonomous Material Design

Abstract

Virtual autonomous material design is a computational paradigm where artificial intelligence, high-throughput simulations, and adaptive optimization work together to discover and refine materials entirely in silico. Models generate candidate structures, simulations evaluate their properties, and closed-loop learning iteratively guides new virtual experiments. This approach reduces cost, accelerates discovery, and uncovers complex structure–property relationships, positioning materials science as a fully computational, data-driven discipline. Achieving this vision requires new methods for representing, generating, and evaluating materials with both physical rigor and computational scalability. Graph theory provides a natural way to encode atomic structures and bonding, enabling machine learning to navigate chemical space while preserving topology and local environments. Group theory complements this by handling symmetry, reducing redundancy and ensuring physically valid configurations. Equally important is automated simulation construction, where workflows translate candidate representations into reliable experiments without human input—an essential step toward true autonomy. This dissertation advances virtual autonomous materials design through integrated developments in mathematical representation, machine learning, and automated simulation. It introduces symmetry-aware, graph-based learning architectures for complex properties, physically grounded methods that eliminate manual tuning in simulations, and hierarchical, group-theoretic approaches for generating chemically meaningful yet tractable design spaces. Building on these foundations, the work couples multi-fidelity prediction, active learning, and high-throughput screening directly to device-level performance, enabling evaluation based on functional impact rather than isolated properties. Together, these contributions establish a closed, co-evolutionary loop in which application demands drive method development, and those methods expand the scope, speed, and realism of autonomous materials discovery.