Prediction of interface structures and energies via virtual screening

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning Applications

Background:

• Material properties are significantly influenced by interfaces due to atomic configuration differences.
• Determining interface atomic structure is crucial but computationally intensive.
• Efficient prediction methods are needed to understand interface properties and facilitate material design.

Purpose of the Study:

• To develop a computationally efficient machine learning-based virtual screening technique for predicting interface energies and structures.
• To demonstrate the robustness and general applicability of the developed method for interface analysis.

Main Methods:

• A virtual screening approach utilizing machine learning was employed.
• Nonlinear regression analysis was performed using geometrical factors (bond length, atomic density) as descriptors.
• Training data from 4 interfaces were used to predict properties for 13 additional interfaces.

Main Results:

• The machine learning method successfully predicted interface structures and energies.
• The developed technique demonstrated a significant efficiency improvement, being hundreds to tens of thousands of times faster than existing methods.
• The use of geometrical factors as descriptors ensures the robustness and general applicability of the method.

Conclusions:

• The developed machine learning-based virtual screening is a powerful and highly efficient tool for determining interface energies and structures.
• This method significantly accelerates materials research by reducing computational costs.
• The approach is expected to advance the understanding of interface nature and facilitate the design of novel material interfaces.