17:15 〜 18:30
[SMP26-P09] The study on stability and properties of N-doped spinel Based on CGCNN and DFT
キーワード：CGCNN, DFT, Spinel
Spinel (AB2X4) is an essential mineral in geology and an important semiconductor and ceramic material. Doping with different atoms will change the properties of spinel and give them unique properties. The crystal graph convolutional neural network (CGCNN) developed by Xie et al. They use graphs to represent the crystal structure and apply neural graph networks to material performance prediction. Taking perovskite as an example, CGCNN has excellent prediction performance.
This study verified whether CGCNN is suitable for predicting the formation energy of doped spinel. The formation energy predicted by CGCNN is compared with the theoretical energy (binding energy) calculated by DFT.
From thousands of CGCNN predicted formation energies of doping Mg2SiO4 structure, which Mg is replaced by Si and O is replaced by N at different atomic ratios. We choose 3 data from the lowest, middle, and highest values respectively (9 data in total). And use DFT to calculate the binding energy at the corresponding structure of doping Mg2SiO4. The calculation results are shown that the changing trend between CGCNN and DFT is almost the same in Figure 1. It implies that CGCNN is suitable for predicting doped spinel. Through the auxiliary exploration of CGCNN. We found a possible nitrogen-doped oxygen structure of spinel (MgAl2O3.5N0.5) in other work.
This work is supported by the Science and Technology Development Fund of Macau (File No. 0111/2020/A, 0042/2018/A2)
This study verified whether CGCNN is suitable for predicting the formation energy of doped spinel. The formation energy predicted by CGCNN is compared with the theoretical energy (binding energy) calculated by DFT.
From thousands of CGCNN predicted formation energies of doping Mg2SiO4 structure, which Mg is replaced by Si and O is replaced by N at different atomic ratios. We choose 3 data from the lowest, middle, and highest values respectively (9 data in total). And use DFT to calculate the binding energy at the corresponding structure of doping Mg2SiO4. The calculation results are shown that the changing trend between CGCNN and DFT is almost the same in Figure 1. It implies that CGCNN is suitable for predicting doped spinel. Through the auxiliary exploration of CGCNN. We found a possible nitrogen-doped oxygen structure of spinel (MgAl2O3.5N0.5) in other work.
This work is supported by the Science and Technology Development Fund of Macau (File No. 0111/2020/A, 0042/2018/A2)