Machine-learning accelerated density-explicit polymer field theory simulations

Area of Science:

• Computational physics
• Polymer science
• Machine learning

Background:

• The density-explicit framework in polymer field theory is gaining traction for its ability to model complex polymer systems.
• This framework offers flexibility in handling various intermolecular potentials and many-body interactions.
• However, simulations using this framework are computationally intensive due to a higher number of fields.

Purpose of the Study:

• To develop a machine learning approach to accelerate polymer field theory simulations.
• To create efficient, low-dimensional feature representations for deep neural networks applicable across different resolutions and dimensions.
• To enable faster and more accurate predictions of polymer phase behavior.

Main Methods:

• Development of deep neural networks (DNNs) with efficient feature representations.
• Application of DNNs to accelerate simulations within the density-explicit framework.
• Testing the DNNs across varying spatial resolutions and dimensions.

Main Results:

• Successfully accelerated polymer field theory simulations using DNNs.
• Demonstrated the effectiveness of low-dimensional and local feature representations.
• Achieved applicability across different spatial resolutions and dimensions.

Conclusions:

• The developed DNNs offer a significant speedup for density-explicit polymer field theory simulations.
• This work provides a foundation for machine learning-assisted tools in polymer and soft matter research.
• Facilitates accurate and efficient prediction of complex block copolymer mesophases.