About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Hume-Rothery Symposium: Interface Structure and Properties: Impact on Microstructure Evolution
|
| Presentation Title |
Reinforcement Learning-Guided Long-Timescale Atomistic Simulations |
| Author(s) |
Ju Li |
| On-Site Speaker (Planned) |
Ju Li |
| Abstract Scope |
Long-timescale processes pose significant challenges in atomistic simulations. We present a computational framework that simulates atomic diffusive relaxation over extended timescales by learning the mean first passage time (MFPT) using a deep neural network. The model is trained with a recursive formalism that emphasizes residence times between neighboring states. Deep reinforcement learning expedites the identification of thermodynamic target states and enables accurate kinetic simulations to estimate atomic transition rates. Applied to vacancy-mediated short-range order (SRO) evolution in equiatomic CrCoNi, the framework resolves the physical timescales of disorder-to-order transitions at various temperatures. [Adv. Sci. 11 (2024) 2304122; arXiv:2411.17839] |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, |