Database Creation, Visualization, and Statistical Learning for Polymer Li+-Electrolyte Design

Visualization and statistical regression of compiled data sets are emerging as powerful tools in understanding and screening the design space of materials properties, rapidly providing insights that would not be readily gained from studies of individual systems. Here, we describe the curation and analysis of a database of polymer Li+-electrolyte conductivity performance, manually extracted from the published literature. We focus on solid, dry polymer electrolytes without additives. Data were extracted from 65 publications, resulting in 655 unique polymer–anion–salt concentration entries and 5225 individual conductivity data points to create an interactive database: PEDatamine.org. Visualization of the collective data set suggested that individual features, other than the activation energy, are poor predictors of conductivity performance across the wide range of polymer chemistries, Li salts, and salt concentrations examined. The Meyer–Neldel rule suggesting a correlation between the conductivity prefactor and activation energy is shown to hold universally for both Arrhenius and Vogel–Fulcher–Tammann representations. Statistical regression techniques were employed to extract the most important features relevant in determining Li+-ion conductivity. These include polymer molecular weight, glass-transition temperature, existence of electronegative heteroatoms in the monomer, and anion size. However, experimental features can be omitted from the regression model without impacting predictive performance, reinforcing the importance of monomer electronegativity, hydrogen bonding, and anion molecular bulk.