Fig. 4: MD validation.
From: De novo design of polymer electrolytes using GPT-based and diffusion-based generative models
a Schematic illustrating the MD simulation workflow. The process begins with the creation of an amorphous polymer-salt system. Subsequent steps include system relaxation and equilibration, followed by the production run. Ion transport properties are then determined using the cluster Nernst-Einstein method based on data collected during the production phase. b Distribution of conductivity and transference number calculated using MD simulations for the top 46 candidates in the generated set (100K samples in total) in comparison to the training data. c box-and-whisker plot showing the conductivity values for each candidate. There are 17 out of 46 candidates showing better conductivity than the optimal polymer (conductivity = 5.1 × 10-4 S/cm) in the train set.
