Multiscale Modeling from Atomistic Approaches with Machine Learning and Advanced Sampling

GS Jung

Abstract: Multiscale modeling methods are typically envisioned as precise and predictive simulation tools to solve complex science and engineering problems. However, even conventional atomistic models are often not sufficient in terms of computational efficiency and accuracy to provide reliable information for the large-scale continuum models. In this seminar, I will focus on code and method developments to overcome current critical limitations. 

In the beginning of the talk, I will showcase studies for crystal growth and failure in 2Dimensional (2D) materials using empirical reactive forcefield (FF). While the models can provide useful insight at the atomic scale, generally developing reliable FFs is extremely limited due to the fixed potential expressions. Recently, neural network (NN) potentials have emerged to overcome such long-standing limitations of empirical potentials.

In the second part, I will present recent developments integrating a density-functional tight-binding code and a PyTorch implementation of NN potentials (TorchANAKIN-ME) into the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) molecular dynamics (MD) software. I will discuss the pros and cons of NN potentials illustrated by a simple carbon system, graphene. While NN potentials can provide higher accuracy than other FFs, e.g., Reactive FF and Adaptive Intermolecular Reactive Empirical Bond Order, and lower computational cost than quantum calculations, efficient sampling or data reduction arises as a critical issue.

In the end, I will present the ongoing development of extended ensemble methods in LAMMPS as an efficient sampling method for different phases and transitions between them. This method can overcome the local-minima problem of conventional MD and predict thermodynamics properties of different phases from atomic simulations, providing missing information for continuum models, e.g., phase-field models, when designing new materials and processes.

The developed models will provide a fundamental understanding of the chemical process and mechanistic insights into the predictive design and interpretive simulation of materials properties and processes relevant to the Oak Ridge National Laboratory’s (ORNL) research programs.

Speaker’s Bio: GS Jung is a Wigner Fellow at ORNL. His research interests are on the multiscale modeling of materials to understand their fundamental properties from synthesis and growth to performance and failures. Before joining ORNL, he earned his Ph.D. in multiscale modeling for 2D materials from the Massachusetts Institute of Technology.


Last Updated: August 5, 2021 - 9:10 am