Machine learning valence force field model

TL;DR

This paper introduces a hybrid machine learning valence force field model that combines physical interpretability with numerical flexibility, enabling accurate and efficient predictions of atomic interactions in materials like graphene.

Contribution

The paper presents the ML-VFF model, integrating the valence force field approach with Gaussian regression to improve prediction accuracy without explicit functional forms.

Findings

Accurately predicts atomic forces and potentials in graphene

Operates with lower computational costs than traditional methods

Highlights advantages and limitations of the ML-VFF approach

Abstract

The valence force field (VFF) model is a concise physical interpretation of the atomic interaction in terms of the bond and angle variations in the explicit quadratic functional form, while the machine learning (ML) method is a flexible numerical approach to make predictions based on some pre-obtained training data without the need of any explicit functions. We propose a so-called ML-VFF model, by combining the clear physical essence of the VFF model and the numerical flexibility of the ML method. Instead of imposing any explicit functional forms for the atomic interaction, the ML-VFF model predicts the potential and force with the Gaussian regression approach. We take graphene as an example to illustrate the ability of the ML-VFF model to make accurate predictions with relatively low computational expenses. We also discuss some key advantages and drawbacks of the ML-VFF model.