Exploring High Thermal Conductivity Amorphous Polymers Using Reinforcement Learning.

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Chemistry

Background:

• Amorphous polymers are crucial for applications requiring efficient thermal transport.
• Traditional material development methods are time-consuming and often unsuccessful.
• Achieving desirable thermal conductivity in polymers remains a significant challenge.

Purpose of the Study:

• To develop a novel inverse design scheme for creating amorphous polymers with enhanced thermal conductivity.
• To utilize reinforcement learning and machine learning for accelerated material discovery.
• To identify polymers with thermal conductivity exceeding 0.400 W/m·K.

Main Methods:

• Employed a reinforcement learning scheme for polymer design.
• Utilized a machine learning model trained on molecular dynamics (MD) simulation data for thermal conductivity prediction.
• Developed a recurrent neural network trained on a large dataset of virtual polymer structures for polymer generation.
• Assessed synthesizability using synthetic accessibility scores and validated thermal conductivity via MD simulations.

Main Results:

• Successfully designed polymers with thermal conductivity above 0.400 W/m·K.
• The best-designed polymer exhibited an MD-calculated thermal conductivity of 0.693 W/m·K.
• The designed polymers were predicted to be easily synthesizable.
• The AI-driven approach significantly outperformed conventional methods.

Conclusions:

• The demonstrated reinforcement learning-based inverse design scheme effectively accelerates the development of amorphous polymers with targeted thermal properties.
• This methodology offers a promising pathway for advancing polymer science and can be extended to other material development tasks.
• The findings pave the way for creating next-generation materials for thermal management solutions.