About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Discovering Continuum Scale Models from Atomistic Simulations |
| Author(s) |
Amit Samanta, Prashanth Ravichandar, John Klepeis |
| On-Site Speaker (Planned) |
Amit Samanta |
| Abstract Scope |
Advancements in high performance computing capabilities have enabled simulations of systems containing billions of atoms thereby providing atomistic insights into macroscopic processes, such as solid-liquid interface instabilities, under extreme non-equilibrium conditions for which continuum scale models are not known or existing continuum scale models breakdown. However, such simulations suffer from the time-scale limitations of molecular dynamics (MD). We present a physics-inspired and data-driven framework to discover partial differential equations (PDEs) for predictive phase field (PF) and PF crystal models from atomistic data. Data from MD simulations of solidification and spinodal decomposition are first mapped to scalar fields, and spatial and temporal derivatives of such fields are used to generate a dictionary. Next, sparse linear and nonlinear regression methods are used with the dictionary to discover a PDE. The discovered PDEs are validated by comparing with MD. Our results show that data-driven frameworks are powerful tools for modeling complex systems. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, Phase Transformations |