Accelerating amorphous polymer electrolyte screening by learning to reduce errors in molecular dynamics simulated...

Area of Science:

• Materials Science
• Computational Chemistry
• Energy Storage

Background:

• Polymer electrolytes are key for advanced lithium-ion batteries.
• Current screening methods using molecular dynamics (MD) simulations are computationally expensive for amorphous polymers.
• Challenges include noise reduction and long simulation times for convergence.

Purpose of the Study:

• To accelerate the large-scale screening of polymer electrolytes.
• To overcome the computational limitations of traditional MD simulations.
• To identify design principles for improved polymer electrolytes.

Main Methods:

• Developed a multi-task graph neural network (GNN).
• Trained the GNN on a combination of short, noisy MD data and long, converged MD data.
• Applied the GNN to screen a large chemical space of 6247 polymers.

Main Results:

• Achieved accurate predictions for 4 converged properties of polymer electrolytes.
• Successfully screened a significantly larger polymer space than previous studies.
• Extracted key design principles for polymer electrolyte development.

Conclusions:

• The GNN approach dramatically accelerates materials discovery for polymer electrolytes.
• This method enables efficient screening of complex amorphous materials.
• An open dataset and design principles are provided for the research community.