| About this Abstract |
| Meeting |
2023 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithm Development in Materials Science and Engineering
|
| Presentation Title |
Enabling Long Timescale Molecular Dynamics Simulation with ab initio Precision |
| Author(s) |
Jan Janssen, Danny Perez |
| On-Site Speaker (Planned) |
Jan Janssen |
| Abstract Scope |
Classical molecular dynamics (MD) is in principle an ideal tool to investigate the long-time evolution of materials, as ab initio-based MD simulations remain limited to very short time. While modern machine learning MD potentials report errors on the order 1 meV/atom, these errors are only typical of configurations that are similar to those found in the training set, and transferability remains limited. This poses a challenge to the accuracy of long-time MD simulations for two reasons: transition rates are exponentially sensitive to energy barriers, and saddle configurations form a very small subset of the whole configuration space and so are very unlikely to appear in traditional datasets. We propose an automated workflow to develop and validate transferable machine learning potentials for long-time simulations. Starting from an information-entropy optimized training set with over 7 million atomic environments, fitted potentials are benchmarked on a set of transition states to characterize their transferability. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Modeling and Simulation, |