A multiscale generative model to understand disorder in domain boundaries.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Materials Science

Background:

• Identifying structural motifs and assembly rules in synthesized materials is challenging, especially with limited atomic resolution microscopy data.
• Understanding domain boundaries is crucial for predicting and optimizing material properties.

Purpose of the Study:

• To propose and validate a hybrid generative model for predicting unseen domain boundaries in potassium sodium niobate thin films.
• To understand the origin of complex domain boundary structures from simple local rules.

Main Methods:

• Utilized a hybrid generative model trained on a small number of observations.
• Predicted domain boundaries without first-principles calculations or atomistic simulations.
• Analyzed domain boundary structures and their potential impact on material properties.

Main Results:

• Successfully predicted unseen domain boundaries in potassium sodium niobate thin films.
• Demonstrated that complex nanometer-scale domain boundaries arise from simple, probabilistic local rules.
• Discovered novel, tileable boundary motifs potentially influencing piezoelectric response.
• Observed domain boundaries with high configurational entropy.

Conclusions:

• Simple, interpretable machine learning models can describe and understand disorder in complex materials.
• The proposed model effectively predicts domain boundaries, offering insights into their formation.
• This work advances functional materials design by elucidating the origin of structural complexity.