About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Search for New Magnetic Material Candidates by First-Principles Calculation and Machine Learning |
| Author(s) |
Sayaka Ito, Insung Seo, Yoshihiro Gohda |
| On-Site Speaker (Planned) |
Sayaka Ito |
| Abstract Scope |
The discovery of new rare-earth permanent magnets superior to Nd2Fe14B is crucial for advancing technologies such as electric motors by enabling enhanced performance and miniaturization. To this end, we are trying to construct two hybrid crystal-structure prediction workflows that combine generic algorithm (GA) with machine learning (ML) models for Nd-based magnet systems. In the first workflow, USPEX is coupled with ALIGNN (NIST, 2021), which uses a line-graph representation of bond angles to capture geometric features and boost the accuracy of formation energy and magnetization predictions. The other, USPEX pairs with CHGNet (LBNL, 2023), which explicitly incorporates magnetic moments to model local charge and spin states. Using a pre-assembled dataset of 40,000 Nd–Fe configurations, we apply both GA–ML workflows to predict crystal structures and identify promising new magnet candidates. We will outline our efforts to employ these GA–ML loops in surveying the compositional and configurational spaces for potential magnet systems. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
ICME, Machine Learning, Computational Materials Science & Engineering |