About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Late News Poster Session
|
| Presentation Title |
H-35: Data-Driven Discovery of Stable Lead-Free Perovskites Via Reinforcement Learning |
| Author(s) |
Ulas Can Yazar, Irmak Sargın |
| On-Site Speaker (Planned) |
Ulas Can Yazar |
| Abstract Scope |
Lead-free perovskites have emerged as sustainable alternatives to lead-based counterparts but continue to face challenges in achieving thermodynamic and environmental stability. This study introduces a computational framework that integrates machine learning and reinforcement learning to accelerate the discovery of stable, non-toxic perovskite compositions. A curated dataset is augmented with elemental descriptors, that are found to be meaningful by principal component analysis (PCA). Supervised models are trained to predict viable A-, B-, and X-site combinations within the ABX₃ lattice, while reinforcement learning dynamically optimizes exploration of compositional space based on model-derived rewards. This workflow enables efficient computational screening of large chemical design spaces, guiding experimental efforts toward the most promising lead-free candidates for photovoltaic applications. |
| Proceedings Inclusion? |
Undecided |
| Keywords |
Machine Learning, Computational Materials Science & Engineering, |