IRG1-3

2021 IRG-1: Bayesian Optimization & Graph Deep Learning for New Materials Discovery

We designed a novel materials screening and discovery workflow that combines Bayesian optimization, graph deep learning (MEGNet model) and density functional theory (DFT) calculations. In the search of hard materials in the Mo-W-Os-Re-B-C chemical space, the workflow reduces the potential candidates from ~400k to just 8, greatly reducing the experimental efforts.

Experiments were carried out on the proposed 8 candidates, and we confirmed two novel hard materials.

The measured hardness values on the newly discovered materials match with our deep learning and DFT predictions, further validating our the materials discovery workflow.

Zuo Y, Qin M, Chen C, Ye W, Li X, Luo J, Ong SP. “Accelerating materials discovery with Bayesian optimization and graph deep learning” Materials Today, In Press (2021). https://doi.org/10.1016/j.mattod.2021.08.012

Jian Luo, Shyue Ping Ong (University of California, San Diego)