About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Data driven discovery of phase field crystal models from atomistic simulations |
| Author(s) |
Amit Samanta |
| On-Site Speaker (Planned) |
Amit Samanta |
| Abstract Scope |
High-performance computing now enables molecular dynamics (MD) simulations of million-atom systems, offering atomistic insight into macroscopic processes under far-from-equilibrium conditions where continuum models are unknown or break down. Yet MD remains fundamentally limited by short accessible timescales. To address this gap, I will present a physics-inspired, data-driven framework that discovers closed-form phase-field-crystal models directly from MD data. Since the discovered models are analytic, they are interpretable, enable discovery of new physics, and can extrapolate beyond the training regime. I will discuss key challenges and machine-learning strategies for: (a) mapping discrete atoms to scalar fields; (b) constructing an expressive dictionary; and (c) ensuring stability of discovered equations. Validation against MD simulations of nanoparticle coalescence demonstrates that data-driven discovery yields predictive and physically grounded continuum models.
This work was performed under the auspices of the U.S. DOE by LLNL under Contract DE-AC52-07NA27344 and funded by LDRD under project tracking code 25-ER-002. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Modeling and Simulation, Computational Materials Science & Engineering |