Standard Deviation Effect of Average Structure Descriptor on Grain Boundary Energy Prediction

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• Grain boundaries (GBs) significantly influence the properties of polycrystalline metals and alloys.
• Understanding GB structure-property relationships is crucial for materials design.
• Existing descriptors may not fully capture the complexity of GB structures.

Purpose of the Study:

• To develop a novel GB structure descriptor for improved energy predictions.
• To investigate the impact of PCF standard deviation on GB energy prediction accuracy.
• To evaluate the performance of the proposed descriptor with different machine learning methods.

Main Methods:

• Development of a new GB structure descriptor by linearly combining average and standard deviation of the two-point correlation function (PCF) with a weight parameter.
• Application of the descriptor to asymmetric tilt GBs in Cu, Al, and Ni (Σ3, Σ5, Σ9, Σ11, Σ13, Σ17).
• Utilizing two machine learning (ML) methods: principal component analysis (PCA)-based linear regression and recurrent neural networks (RNN).

Main Results:

• The proposed GB structure descriptor significantly improves GB energy prediction accuracy for both PCA-based linear regression and RNN models.
• Incorporating the standard deviation of PCF enhances the descriptor's ability to differentiate between various GB structures.
• The method of GB atom selection for PCF evaluation was found to impact prediction outcomes.

Conclusions:

• The novel GB structure descriptor, incorporating data dispersion information, offers superior predictive power for GB energy.
• This descriptor enhances the discriminatory capability beyond using only the average PCF.
• Further research into GB atom selection methodologies is recommended for optimizing PCF-based predictions.