Toward Predicting Intermetallics Surface Properties with High-Throughput DFT and Convolutional Neural Networks.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid State Physics

Background:

• Surface energy is critical for material properties and design, but predicting it for complex alloys remains challenging.
• Cleavage energy computation is a key step towards calculating surface energy for extensive material exploration.
• Existing methods are limited for predicting surface energies of bimetallic and more complex crystalline structures.

Purpose of the Study:

• To develop a predictive workflow for calculating cleavage energies of intermetallic alloys using high-throughput density functional theory (DFT) and machine learning.
• To create a comprehensive database of cleavage energies for a wide range of alloys.
• To enable efficient screening of stable surfaces for novel materials discovery.

Main Methods:

• Implemented a high-throughput DFT workflow to calculate cleavage energies for 3033 intermetallic alloys.
• Compiled a database of calculated cleavage energies.
• Trained a crystal graph convolutional neural network (CGCNN) model on the generated dataset.

Main Results:

• The CGCNN model achieved high accuracy in predicting cleavage energy, with a mean absolute test error of 0.0071 eV/Ų.
• The workflow successfully predicted stable surfaces and qualitatively reproduced nanoparticle surface distributions (Wulff constructions).
• Generated quantitative insights into unexplored chemical spaces for materials stability.

Conclusions:

• The developed workflow provides a powerful tool for predicting cleavage energies and identifying stable surfaces in unexplored chemical spaces.
• This approach facilitates the down-selection of promising candidate materials for applications like catalyst screening and nanomaterial synthesis.
• The study advances the design and discovery of inorganic crystals with tailored surface properties.