Automated pyiron workflows accelerate the development of machine learning potentials in materials science (NFDI-MatWerk)

The implementation of pyiron-based automated workflows fundamentally improves the creation of machine learning interatomic potentials (MLPs) in materials science. This comprehensive and user-friendly framework, built upon the pyiron integrated development environment (IDE), allows researchers to seamlessly navigate the entire MLP development cycle. The process begins with the creation of systematic Density Functional Theory (DFT) databases, followed by fitting the DFT data to various empirical potentials or MLPs, and culminates in the validation of these potentials through a largely automated approach. The framework’s power is exemplified through its application to three distinct classes of interatomic potentials: the embedded atom method (EAM), high-dimensional neural network potentials (HDNNP), and atomic cluster expansion (ACE). Each method showcases the versatility and efficiency of the pyiron workflows. A notable success story involves the computation of a binary composition-temperature phase diagram for the Al-Li alloy, a lightweight material critical to the aerospace industry. This advanced validation not only demonstrates the framework’s capabilities but also highlights its potential to accelerate research and development in materials engineering, paving the way for innovative applications in various technological fields. A more in-depth description of successful pyiron implementation can be found in the Nature Nature Communications Article.

More information:

• https://www.nature.com/articles/s41524-024-01441-0
• https://pyiron.org/

Other posts

Posts