| Abstract Scope |
Computational discovery of materials requires representations that connect electronic/atomistic theory, simulation, and experiment. I will present three physics-grounded AI workflows. First, a primitive-modulation representation describes commensurate and incommensurate magnetic structures within the chemical unit cell, enabling prediction of non-collinear magnetic order from atomic coordinates. Second, domain-adaptive machine learning bridges stochastic models of grain-boundary dynamics with X-ray photon correlation spectroscopy, allowing physical parameters to be extracted from noisy non-equilibrium experiments. Third, using high-temperature superconductivity as a case study, I will show how large language models can organize decades of literature into mechanism-level correlations and trends, revealing latent scientific knowledge beyond any single paper. These examples suggest a path where AI accelerates materials discovery not by replacing physics, but by embedding physical representations, simulations, and experimental validation into the workflow. |