Session: 04-28-02: Modeling and Experiments in Nanomechanics and Nanomaterials

Paper Number: 113427

113427 - Machine Learning Accelerated Atomistic Simulations for 2D Materials With Defects 

2D materials generally show very different physical and chemical properties from 3D materials, which provide them with promising applications in cutting-edge technology areas like electronics, catalysis, biosensing and nanogenerators. Among the many amazing and unique properties of 2D materials, mechanical properties play a significant role in the performance of their potential applications. Although some of the 2D materials are reportedly characterized by experimental methods such as nanoindentation, it is extremely hard for a good experimental control, and some results are counter-intuitive and highly controversial. To better understand and illustrate their unique properties, current research heavily relays on atomistic simulations, while successful molecular dynamics (MD) simulations require the high reliability of interatomic interaction potentials that empirical potentials usually cannot provide. On the other hand, the ab initio calculations, for example density functional theory (DFT), are able to conduct high-fidelity simulations in essence, but with a high computational cost and largely limited simulation size. Recently, machine learning potentials (MLPs) become a trend to interpolate the potential energy surface based on artificial neural networks (ANNs) with reference datasets generated from first principal calculations. It is known that machine learning potentials have been developed and verified for many material systems including 2D materials, but there is limited work published regarding structural defects. However, the presence of defects, especially in 2D materials is inevitable, making the mechanical properties susceptible to degradation, so the possible profound effect is worth studying. Using graphene as a model material, here we demonstrate a new machine learning potential for 2D materials with defects. First the reference datasets are generated by DFT calculations using Quantum Espresso (QE) software. Second the atomic structures are coded using perturbation-invariant representations known as symmetry functions. Ten radial and twenty angular symmetry functions are used in this work and the parameters are gathered from published literature. Then the ANN with two hidden layers is used to train the potential, which has been demonstrated to be suitable for developing machine learning potential in our previous work. Finally the trained potential is imported into LAMMPS to conduct MD simulations and the results are verified with DFT results. We show that only a small size of training dataset can yield a good machine learning potential and provide accurate cohesive energy predictions on defective graphene. It is expected that our work provides fundamental support on investigating defective 2D materials and understanding defect effects to develop structure-property relationship. This will also promote the development of machine learning based simulation tools for the study and design of complex materials.

Presenting Author: Shijie Sun University of Illinois at Urbana Champaign

Presenting Author Biography: Shijie is a Ph.D student at UIUC.

Authors:

Shijie Sun University of Illinois at Urbana Champaign
Akash Singh University of Illinois at Urbana Champaign
Yumeng Li University of Illinois at Urbana-Champaign