Bayesian optimization based on element mapping to design high-capacity NASICON-type cathode in sodium-ion battery

Abstract

To discover novel materials with high performance, there have been many attempts to adopt Bayesian optimization (BO) to materials science, owing to its efficiency in navigating complex and high-dimensional design spaces. However, the application of BO to material design has been suffered from handling discrete input variables, such as elements. Here, we introduce a novel element mapping strategy that encodes elemental identities into chemically meaningful continuous values, enabling to create easy-to-predict chemical spaces. We apply this new framework to design high capacity Na3V2(PO4)2F3 (NVPF) cathode materials for sodium-ion batteries, targeting that shift all working voltages into the desired operational voltage window. The proposed framework successfully suggests 16 optimal element composition within 50 iterations. Our results demonstrate the way to overcome the limitation of categorical input that will likely broaden the applicability of BO to a wider range of material discoveries.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.