Enhancing Thermal Conductivity Computation of Polymers via Machine Learning Techniques.

Area of Science:

• Materials Science
• Computational Chemistry
• Polymer Physics

Background:

• Accurate prediction of polymer thermal conductivity (κ) is challenging due to complex structures.
• Existing ab initio methods (e.g., DFT-BTE) are computationally expensive.
• Classical force fields in molecular dynamics lack the necessary accuracy for κ prediction.

Purpose of the Study:

• To develop a computationally efficient and accurate method for predicting polymer thermal conductivity.
• To enable quantitative prediction of bulk polymer κ from small quantum cluster data.
• To validate the developed method against experimental data.

Main Methods:

• Combining ab initio hybrid machine learning (ML) with multipolar polarizable potentials (PhyNEO).
• Utilizing ML-facilitated heat flux calculation for reliable trajectory generation.
• Employing poly(ethylene oxide) as a model system for validation.

Main Results:

• The PhyNEO-ML approach provides reliable heat flux trajectories.
• Quantitative prediction of polymer κ was achieved with excellent agreement.
• Calculated results for poly(ethylene oxide) matched experimental data from time-domain thermoreflectance measurements.

Conclusions:

• The developed hybrid ML/PhyNEO method offers a computationally feasible and accurate approach for polymer thermal conductivity prediction.
• This method allows for quantitative κ predictions starting from minimal quantum data.
• The approach shows broad applicability for future polymer material design and analysis.