Huan Tran¹, Abhirup Patra¹, Deepak Kamal¹, Lihua Chen¹, Chiho Kim¹, Rampi Ramprasad¹; ¹Georgia Institute of Technology

Two primary challenges in developing a computational polymer database are identifying the polymers to be considered and constructing reasonable crystal models for calculations. We have developed a rational scheme, which involves an active-learning guidance step and a polymer crystal structure prediction strategy, to solve these problems. Our scheme has been used to significantly enlarge a computational polymer database in a guided manner, promoting the diversity of the data in both the chemical space and the property space (taking polymer band gap as an example). Using the obtained data, we have developed a powerful multi-fidelity co-Kriging model for accurately and rapidly predicting polymer band gap. This model is available in the Polymer Genome online platform (https://www.polymergenome.org/).