Λεπτομέρειες βιβλιογραφικής εγγραφής
| Τίτλος: |
TransMagNet: prediction of crystal system and space group for crystalline materials based on composition using deep learning. |
| Συγγραφείς: |
Xue, Jihang1 (AUTHOR), Luo, Tianjun1 (AUTHOR), Jiang, Yongquan1,2 (AUTHOR) yqjiang@swjtu.edu.cn, Yang, Yan1,2 (AUTHOR), Gong, Kuanpin1 (AUTHOR), Deng, Zigang1 (AUTHOR), Feng, Qingguo1 (AUTHOR), Zhang, Weihua1 (AUTHOR) |
| Πηγή: |
Journal of Applied Crystallography. Dec2025, Vol. 58 Issue 6, p1870-1879. 10p. |
| Θεματικοί όροι: |
*Space groups, *Deep learning, *Machine learning, *Structural analysis (Science), *Crystal structure, *Crystals, *Statistical reliability |
| Περίληψη: |
The prediction of crystal systems and space groups has been widely used in the estimation of crystalline material properties and structure prediction. Previous research regarding structure determination methods based on X‐ray diffraction experiments and density functional theory has attained remarkable efficacy and performance, but these approaches are not applicable to large‐scale screening. There are also machine learning models that use Magpie descriptors for space group prediction; for example, that proposed by Liang et al. [Phys. Rev. Mater. (2020), 4, 123802] exhibited prediction accuracies ranging between 0.638 and 0.907 for different types of crystals. Here we put forward a branch network model, named TransMagNet, which is based on transformer encoders [Vaswani et al. (2017). Advances in neural information processing systems 30, pp. 5998–6008] and Magpie linear layers [Ward et al. (2016). npj Comput. Mater.2, 16028], to predict the crystal systems and space groups of materials by merely relying on compositional data. Benchmarking on the Materials Project dataset demonstrates a significant performance improvement in space group classification of our model over previous models, with an accuracy spanning from 0.811 to 0.981 and a maximum improvement in prediction accuracy of 6.5%. The model also achieves a significant performance improvement in crystal system prediction, with an accuracy of 0.854. [ABSTRACT FROM AUTHOR] |
| Βάση Δεδομένων: |
Academic Search Index |