Large-scale atomistic simulation of dislocation core structure in face-centered cubic metal with Deep Potential method

TL;DR

This paper demonstrates the use of a Deep Potential machine learning model trained on DFT data to simulate dislocation core structures in face-centered cubic copper at a large scale, surpassing traditional computational limitations.

Contribution

It introduces a Deep Potential approach for large-scale, accurate atomistic simulation of dislocation cores in metals, bridging the gap between DFT accuracy and classical methods.

Findings

Deep Potential accurately reproduces dislocation core structures and energies.

The method is validated against a Peierls model.

Extension to dislocations with defects is feasible.

Abstract

The core structure of dislocations is critical to their mobility, cross slip, and other plastic behaviors. Atomistic simulation of the core structure is limited by the size of first-principles density functional theory (DFT) calculation and the accuracy of classical molecular dynamics with empirical interatomic potentials. Here, we utilize a Deep Potential (DP) method learned from DFT calculations to investigate the dislocations of face-centered cubic copper on a large scale and obtain their core structures and energies. The validity of the DP description of the core structure and elastic strain from dislocation is confirmed by a fully discrete Peierls model. Moreover, the DP method can be further extended easily to dislocations with defects such as surface or vacancy, and our study will pave a way in the large-scale atomistic simulation of dislocation on the DFT level.