Toward Designing Highly Conductive Polymer Electrolytes by Machine Learning Assisted Coarse-Grained Molecular Dynamics

TLDR

Solid polymer electrolytes are promising for next‑generation lithium‑ion batteries but suffer from low ionic conductivity. The study proposes a design approach that combines coarse‑grained molecular dynamics with machine learning to develop highly conductive SPEs. The authors construct a high‑dimensional design space of physically interpretable descriptors via coarse‑graining and use Bayesian optimization to autonomously run CGMD simulations for efficient exploration. The CGMD‑BO approach revealed how lithium conductivity depends on molecular size and nonbonding interactions, offering guidance to enhance anion, secondary site, and backbone chain components of top electrolytes.

Abstract

Solid polymer electrolytes (SPEs) are considered promising building blocks of next-generation lithium-ion batteries due to their advantages in safety, cost, and flexibility. However, current SPEs suffer from a low ionic conductivity, motivating the development of novel highly conductive SPE materials. Here we propose a new SPE design approach that integrates coarse-grained molecular dynamics (CGMD) with machine learning. A continuous high-dimensional design space, composed of physically interpretable universal descriptors, was constructed by the coarse graining of chemical species. A Bayesian optimization (BO) algorithm was then employed to efficiently explore this space via autonomous CGMD simulations. Adopting this CGMD-BO approach, we obtained comprehensive descriptions of the relationships between the lithium conductivity and intrinsic material properties at the molecular level, such as the molecule size and nonbonding interaction strength, to provide guidance on directions to improve upon the components of the best-known electrolytes, including anion, secondary site, and backbone chain.