Designing Accurate Moment Tensor Potentials for Phonon-Related Properties of Crystalline Polymers

Area of Science:

• Materials Science
• Computational Chemistry
• Polymer Physics

Background:

• Simulating phonon-related properties of crystalline polymers is crucial for applications but challenging due to system size limitations for ab initio methods and inaccuracies of classical force fields.
• Machine-learned potentials (MLPs) offer a promising solution by combining high accuracy with computational efficiency when parameterized with material-specific data.

Purpose of the Study:

• To develop and validate a protocol for parameterizing moment tensor potentials (MTPs) for crystalline polymers.
• To ensure optimal performance and thorough benchmarking of MTPs for molecular dynamics simulations.
• To enable the accurate simulation of properties previously inaccessible.

Main Methods:

• Devised a protocol for parameterizing MTPs using material-specific ab initio data.
• Trained MTPs for polyethylene (PE), polythiophene (PT), and poly-3-hexylthiophene (P3HT).
• Simulated structural properties, phonon band structures, elastic constants, forces, thermal conductivity, and thermal expansion.

Main Results:

• The developed MTP protocol enables accurate and efficient simulations of crystalline polymers.
• Simulations for PE, PT, and P3HT successfully reproduced structural properties, phonon band structures, elastic constants, and forces.
• For PE, simulated thermal conductivity and thermal expansion showed good agreement with experimental data.

Conclusions:

• The established MTP parametrization protocol significantly advances the simulation capabilities for crystalline polymers.
• This approach allows for a massive speedup in complex calculations while maintaining high accuracy, comparable to DFT.
• Reliable simulation of challenging properties, such as thermal conductivity and expansion, is now achievable.