学术文献库

Machine learning models and applications in materials design and discovery typically involve the use of feature representations or "descriptors" followed by a learning algorithm that maps them to a user-desired property of interest. Most popular mathematical formulation-based descriptors are not unique across atomic environments or suffer from transferability issues across different application domains and/or material classes. In this work, we introduce the Crystal Edge Graph Attention Neural Network (CEGANN) workflow that uses graph attention-based architecture to learn unique feature representations and perform classification of materials across multiple scales (from atomic to mesoscale) and diverse classes ranging from metals, oxides, non-metals and even hierarchical materials such as zeolites and semi ordered materials such as mesophases. We first demonstrate a case study where the classification is based on a global, structure-level representation such as space group and structural dimensionality (e.g., bulk, 2D, clusters etc.). Using representative materials such as polycrystals and zeolites, we next demonstrate the transferability of our network in successfully performing local atom-level classification tasks, such as grain boundary identification and other heterointerfaces. We also demonstrate classification in (thermal) noisy dynamical environments using a representative example of crystal nucleation and growth of a zeolite polymorph from an amorphous synthesis mixture. Finally, we characterize the formation of a binary mesophase and its phase transitions and the growth of ice, demonstrating the performance of CEGANN in systems with thermal noise and compositional diversity. Overall, our approach is agnostic to the material type and allows for multiscale classification of features ranging from atomic-scale crystal structures to heterointerfaces to microscale grain boundaries.