An atomic-scale digital twin of dislocation systems

Dislocations, the primary carriers of plasticity, govern the mechanical properties of crystalline materials ranging from metals to semiconductors. The atomic-scale behavior of a dislocation is controlled by multi‑dimensional fields encompassing structure, stress, energy, and chemical environment, which govern its intrinsic mobility and interactions with other defects. However, simultaneously capturing such multi‑dimensional information remains a long‑standing challenge, since the atomic‑resolution images primarily reveal structural features like atomic positions, while other critical fields remain hidden. Here, we introduce AtomDiT, a machine learning-based framework that creates a digital twin of dislocations directly from a high‑resolution micrograph. AtomDiT uses atomistic simulations to learn the relationships between local atomic environment and multi‑dimensional fields, and transfers this knowledge to experimental images. Derived from a single image, it identifies dislocation types and core positions, constructing the surrounding mechanical and chemical fields. AtomDiT is applicable across various crystal structures, dislocation types, chemical complexities, and deformation states. Using the digital twin, we reveal how stress and elemental distribution regulate dislocation‑interface interactions, offering mechanistic insights into plasticity in chemically and structurally complex systems. Based on atomic structures derived from experimental measurements, the digital twin directly serves as a data foundation for mesoscale and continuum‑level plasticity models, bridging atomic‑scale defect dynamics with higher‑scale material behavior.