Predicting copolymer critical parameters with a theory-integrated neural network.

Area of Science:

• Polymer Science
• Computational Chemistry
• Materials Science

Background:

• Polymer solution phase behavior is crucial for materials design.
• Classical theories like Flory-Huggins have limitations in predicting sequence-specific effects.
• Accurate prediction of phase behavior requires advanced computational approaches.

Purpose of the Study:

• To develop a machine learning framework for predicting copolymer phase behavior.
• To enhance prediction accuracy and interpretability by integrating physical insights into neural networks.
• To identify key determinants of copolymer critical parameters.

Main Methods:

• Developed a theory-integrated neural network (TI-NN) combining neural networks (NNs) with scaling relations.
• Utilized grand canonical Monte Carlo simulations for 3351 model copolymer sequences.
• Analyzed feature importance to understand determinants of phase behavior.

Main Results:

• A standard NN provided reasonable accuracy, but the TI-NN significantly reduced prediction errors.
• The TI-NN demonstrated robust extrapolation capabilities beyond the training dataset.
• Solvent selectivity and sequence blockiness were identified as dominant factors influencing critical parameters.

Conclusions:

• Integrating theoretical insights into machine learning models improves accuracy and interpretability for copolymer phase behavior prediction.
• The developed TI-NN framework offers a powerful tool for designing polymers with targeted properties.
• This approach advances the understanding and prediction of complex polymer solution phenomena.