About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Computational Thermodynamics and Kinetics
|
Presentation Title |
Normalizing Flows for Accelerating Atomistic Simulation of Rare Events |
Author(s) |
Rasool Ahmad, Wei Cai |
On-Site Speaker (Planned) |
Rasool Ahmad |
Abstract Scope |
Hyperdynamics molecular dynamics simulation is a powerful technique to directly study the atomic-scale rare events in solids which control important finite-temperature materials properties. Hyperdynamics MD employs a bias potential to accelerate the rare events whose design has remained one of the challenging problems. In this work, we show that a recently discovered property of normalizing flow, a deep generative model, to learn the underlying potential energy function can be utilized to automate the designing of bias potential to perform the hyperdynamics MD of rare events. We demonstrate the effectiveness of the developed method by applying it to three problems: transition between two metastable states in a two-dimensional potential energy surface, vacancy migration, and adatom diffusion. The results show the promise of normalizing flow to efficiently study the atomic-scale rare event processes. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, Computational Materials Science & Engineering, Modeling and Simulation |