Bayesian Optimization Guided Coarse-Grained Molecular Dynamics for Designing Highly Conductive Solid Polymer Electrolytes

Solid polymer electrolytes (SPEs) are promising building blocks for the next-generation lithium-ion batteries, due to their unique advantages in safety, cost and flexibility. However, the incorporation of current SPEs into real-world applications is largely limited by their low ionic conductivity, which motivates the innovation and development of highly conductive SPE materials. Here we develop a computational materials exploration approach that combines the Bayesian Optimization (BO) algorithm with coarse-grained molecular dynamics (CGMD) simulations. While the CGMD effectively reduces the computational cost in comparison with the fully atomistic models, it reasonably maintains the capability of capturing the ion motions and the polymer conformational evolution. Also, we utilize the CGMD input parameters, including the particle sizes and the intermolecular interaction strengths, as universal and interpretable descriptors, to construct a continuous CG design space that encodes the information from the conventional chemical species space. Adopting the BO algorithm, we conduct efficient explorations of this high-dimensional CG space iteratively, towards achieving complete descriptions of the correlations between the molecular level information and the transport properties. The trained CGMD-BO model predicts the conductivities for several commonly used polymer electrolyte systems, all in good agreement with the experimental measurements. In addition, the model suggests new directions to improve upon existing electrolytes, by showing the dependences of the lithium conductivity on the properties of anions, secondary sites, and backbone chains, respectively. This research improves our understanding of the ion transport mechanisms in polymers, as well as provides a new tool for designing tailored SPE materials.