|
Neural evolution structure generation: High Entropy Alloys
C.G. Tetsassi Feugmo, K. Ryczko, A. Anand, C. Veer Singh, and I. Tamblyn Journal of Chemical Physics, 155, 044102 (2021) - Cover Article Generating realistic atomic configurations for high entropy alloys (HEAs) is a major computational challenge due to the vast combinatorial space of possible element arrangements. In this work, we introduce a neural evolution structure (NES) approach that combines artificial neural networks with evolutionary algorithms to design HEA configurations via an inverse design methodology. By leveraging pair distribution functions and atomic properties, the method can be trained on smaller unit cells and then used to generate much larger structures. The NES approach achieves approximately 1000 times faster computational speed compared to the conventional Special Quasirandom Structures (SQS) method, enabling the generation of structures containing over 40,000 atoms within hours rather than days. Unlike traditional SQS methods, a single trained model can produce multiple distinct structures while maintaining identical fractional compositions, substantially expanding the accessible design space for these complex multi-component alloys. |
