High-Temperature Polymer Dielectrics Designed Using an Invertible Molecular Graph Generative Model.

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Computational material design faces challenges in generating molecules with specific properties.
• Data-driven material informatics, utilizing deep learning, offers a promising avenue for accelerating material discovery.
• High-temperature polymer dielectrics are crucial for advanced energy storage applications.

Purpose of the Study:

• To develop and apply an invertible graph generative model for designing hypothetical high-temperature polymer dielectrics.
• To leverage deep generative models for efficient discovery of materials with desired properties.
• To generate polymers suitable for high-temperature energy storage film capacitors.

Main Methods:

• Trained an invertible normalizing flow-based molecular graph generative model on a dataset of 250k polymer molecular graphs.
• Utilized invertible transformations between latent distributions and molecular graph structures.
• Employed a one-shot generation process involving sampling from a latent space and generating molecular graphs via two invertible flows.

Main Results:

• Successfully generated hypothetical polymer molecular graphs with desired properties.
• The model demonstrated exact likelihood training and efficient generation.
• Identified potential polymer candidates with high glass-transition temperature (Tg) and wide band gap (Eg).

Conclusions:

• The invertible graph generative model effectively facilitates the design of high-temperature polymer dielectrics.
• Deep generative models show significant potential in accelerating the discovery of advanced materials for energy storage.
• This work contributes to the efficient computational design of functional polymers.