Enhanced Simulation of Complicated MXene Materials with Graph Convolutional Neural Networks

Area of Science:

• Materials Science
• Computational Materials Science
• Nanomaterials

Background:

• MXenes, 2D transition metal carbides, show promise in energy storage, sensors, and catalysis.
• Material properties are linked to electronic structure, specifically density of states (DOS).
• Traditional density functional theory (DFT) is computationally expensive for complex compositions like high-entropy MXenes.

Purpose of the Study:

• To apply machine learning (ML) for predicting the DOS of complex high-entropy MXenes.
• To assess the accuracy of ML models in reproducing DOS spectra.
• To utilize ML-predicted DOS for screening potential electrode materials for lithium batteries.

Main Methods:

• Crystal graph convolutional neural networks (CGCNN) model was employed.
• DFT calculations on M3C2 and M4C3 structures served as training data.
• Predicted DOS was used to calculate lithium adsorption energy.

Main Results:

• The CGCNN model accurately reproduced the DOS of high-entropy MXenes based on atomic structure.
• Lithium adsorption energies were precisely predicted using ML-generated DOS.
• The ML approach demonstrated efficiency and accuracy in predicting MXene properties.

Conclusions:

• Machine learning, specifically CGCNN, offers an efficient alternative to DFT for complex MXene systems.
• ML streamlines the prediction of electronic properties and facilitates the discovery of new materials.
• This work enhances the understanding of MXene intrinsic properties and accelerates their application in energy storage.