Screening for novel materials using machine learning
||Screening for novel materials using machine learning|
||Rickard Armiento <email@example.com>|
||2023-03-03 – 2023-10-01|
Data from various materials databases play a crucial role in developing accurate ML models for materials discovery. It is observed that many of the entries in these databases are seeded from experiment data, which puts an upper bound on the structural diversity of materials in these databases. Our research seeks to uncover previously unseen crystal structures by coarse graining the space by describing crystal structures based on their symmetries.
This project is a continuation of Berzelius-2022-192 where we benchmarked and optimized our GPU software and workflows for screening novel materials using the machine learning model in https://doi.org/10.1126/sciadv.abn4117. The allocation allowed us to improve the software and it is now ready for trial runs.
This proposal is for an allocation to start the first trial runs of a screening of crystal structures for materials predicted to be stable via their symmetries. We will use the computational resources to evaluate the accuracy and power of these methods in preparation of a possible future larger-scale screening effort.