First-Principles-Based Machine-Learning Molecular Dynamics for Crystalline Polymers with van der Waals Interactions.

Area of Science:

• Computational Materials Science
• Polymer Physics
• Machine Learning Applications

Background:

• Machine-learning (ML) techniques are increasingly vital for high-throughput screening and predicting material properties across scales.
• ML force fields (FFs) with quantum mechanical accuracy are crucial for advanced material simulations.
• Developing ML-FFs for polymers is challenging due to their complex atomic configurations.

Purpose of the Study:

• To demonstrate the effective development of ML-FFs for an organic polymer, polytetrafluoroethylene (PTFE).
• To utilize kernel functions and Gaussian processes for constructing accurate ML-FFs.
• To validate the ML-FF's ability to predict physical properties of PTFE.

Main Methods:

• Acquired a dataset using first-principles calculations and ab initio molecular dynamics (AIMD) simulations.
• Developed ML-FFs for PTFE using kernel functions and a Gaussian process.
• Validated ML-FF performance against density functional theory (DFT) calculations for various chain lengths.

Main Results:

• Optimized structures of longer PTFE chains using the ML-FF showed excellent consistency with DFT calculations, despite training on short chains.
• The ML-FF successfully described key physical properties of a PTFE bundle, including density, melting temperature, coefficient of thermal expansion, and Young's modulus, when integrated with molecular dynamics simulations.

Conclusions:

• The developed ML-FF provides a computationally efficient and accurate method for simulating PTFE properties.
• This approach overcomes the limitations of traditional methods in handling the complex configurational space of polymers.
• The study highlights the potential of ML-FFs for advancing polymer science and materials discovery.