Machine learning of carbon vacancy formation energy in high-entropy carbides. Issue 4 (April 2023)

• Record Type: Journal Article
• Authors: Zhao, Xi; Yu, Sen; Zheng, Jiming; Reece, Michael J.; Zhang, Rui-Zhi
• Is Part Of: Journal of the European Ceramic Society. Volume 43:Issue 4(2023)
• Page Start: 1315
• Page End: 1321
• Publication Date: 2023-04
• Subjects: High-entropy carbides (HECs) -- High-throughput density functional theory calculation (HT-DFT) -- Machine Learning (ML) -- Vacancy; Ceramic materials -- Periodicals; Composite materials -- Periodicals; Matériaux céramiques -- Périodiques; Composites -- Périodiques; Ceramic materials; Composite materials; Periodicals; Electronic journals; 666.05
• DOI: 10.1016/j.jeurceramsoc.2022.11.044
• Languages: English
• ISSNs: 0955-2219

Abstract

Abstract: Carbon vacancies as intrinsic defects in high-entropy carbides (HECs) strongly affect their physical and chemical properties, it is important to understand the details of vacancy formation at an atomic level. In this work, 1280 carbon vacancies are investigated via high-throughput density functional theory (HT-DFT) calculations and machine learning (ML) models on their formation energies. We found that more than half of the carbon vacancies are preferentially surrounded by three types of cations, with Hf less likely to be present. The role of metal cations in promoting the formation of carbon vacancies is: Nb＞Ta＞Zr＞Hf, indicating the possibility to modify the carbon vacancy concentration by change the cation elemental ratios. A random forest model was trained on DFT results to predict vacancy formation energies with a mean absolute error of 54 meV. The feature importance analysis is consistent with the analysis of the DFT results.