Gaming self-consistent field theory: Generative block polymer phase discovery

Area of Science:

• Soft Matter Physics
• Polymer Science
• Computational Materials Science

Background:

• Block polymers are model systems for studying soft matter self-assembly due to their describable thermodynamics via self-consistent field theory (SCFT).
• SCFT is effective for explaining experimental data and constructing phase diagrams, leading to interest in its use for guiding experimental design.
• A key limitation of SCFT for phase discovery is the requirement to pre-specify candidate phases, hindering the identification of novel morphologies.

Purpose of the Study:

• To overcome the challenge of discovering new morphologies in block polymers using SCFT.
• To develop a computational pipeline for accelerated block polymer phase discovery.
• To demonstrate a novel approach for generating candidate phases beyond known structures.

Main Methods:

• Trained a deep convolutional generative adversarial network (GAN) using trajectories from converged SCFT solutions.
• Deployed the trained GAN to generate input fields for subsequent SCFT calculations.
• Applied the computational pipeline to investigate network phase formation in neat diblock copolymer melts.

Main Results:

• The GAN successfully generated a diverse set of 349 candidate phases from a small training set of five networks.
• The discovered phases included both known morphologies and previously unexplored structures.
• A novel chiral network morphology was identified, showcasing the method's capability for discovering complex structures.

Conclusions:

• The developed computational pipeline, integrating GANs and SCFT, effectively addresses the challenge of block polymer phase discovery.
• This AI-driven approach significantly expands the exploration of soft matter morphologies.
• The open-source pipeline holds promise for widespread application in block polymer research and other soft matter systems.