Random Forest Predictor for Diblock Copolymer Phase Behavior.

Area of Science:

• Polymer Science
• Materials Science
• Computational Chemistry

Background:

• Physics-based models like self-consistent field theory (SCFT) are standard for predicting block copolymer phase behavior.
• SCFT requires accurate Flory-Huggins interaction parameters (χAB) and Kuhn length ratios (bA/bB), which are challenging to measure.
• These parameters quantify block incompatibility and chain dimensions, crucial for material design.

Purpose of the Study:

• To develop a machine learning model for predicting AB diblock copolymer phase behavior.
• To bypass the need for experimental measurement of χAB and bA/bB parameters.
• To improve the accuracy and accessibility of phase behavior prediction for material design.

Main Methods:

• Utilized a random forest machine learning approach.
• Trained the model on a dataset of 4768 data points derived from chemical identities of polymer blocks.
• Evaluated model performance against SCFT using group contribution theory for χAB.

Main Results:

• The random forest model achieved nearly 90% accuracy in predicting phase behavior.
• This accuracy is approximately double that of SCFT using group contribution theory.
• The model maintained at least 60% accuracy even with highly uncertain molecular parameter measurements.

Conclusions:

• Machine learning, specifically random forest, provides a highly accurate method for predicting block copolymer self-assembly.
• This approach eliminates the need for laborious measurement of molecular parameters like χAB and bA/bB.
• The model shows potential for accelerating the design and discovery of new self-assembling materials.