Deep-Learning Interatomic Potential Connects Molecular Structural Ordering to the Macroscale Properties of...

Area of Science:

• Polymer Science
• Computational Chemistry
• Materials Science

Background:

• Polyacrylonitrile (PAN) is a vital commercial polymer with atactic stereochemistry.
• Understanding PAN's molecular interactions is crucial for optimizing product design and reducing processing costs.
• Traditional ab initio molecular dynamics (AIMD) are accurate but limited to small molecules, hindering large-scale polymer analysis.

Purpose of the Study:

• To develop a scalable computational method for analyzing large-scale polymer structures and properties.
• To investigate the influence of molecular interactions on PAN's bulk structure and mechanical properties.
• To establish accurate structure-property relationships for PAN and related polymers.

Main Methods:

• Training neural network interatomic potentials (NNIPs) on small-scale AIMD data of PAN oligomers.
• Employing NNIPs for large-scale simulations of amorphous bulk PAN.
• Validating NNIP-predicted structures against experimental X-ray structure factor data.
• Comparing predicted properties (density, elastic modulus) with experimental values.

Main Results:

• NNIPs successfully predict the amorphous bulk PAN structure, capturing intra- and interchain hydrogen-bonding and dipolar correlations.
• Predicted density and elastic modulus align well with experimental data.
• A strong correlation was observed between elastic modulus and PAN structural orientation (Hermans orientation factor).

Conclusions:

• NNIPs offer a computationally efficient and accurate method for simulating large-scale polymer systems.
• This approach provides critical insights into polymer structure-property relationships.
• The study paves the way for sustainable, ab initio accuracy in predicting properties of PAN and analogous polymers across scales.