Paper Image

Machine learning for molecular dynamics simulations

Published on:

10 January 2023

Primary Category:

Machine Learning

Paper Authors:

Shaswat Mohanty,

Sanghyuk Yoo,

Keonwook Kang,

Wei Cai

Bullets

Key Details

Machine learning models can predict forces for molecular dynamics

Graph neural networks are a promising architecture

Liquid properties are reproduced when training only uses liquid data

Solid properties like phonon frequencies require solid and liquid training data

Comprehensive benchmarking is critical when developing these models

AI generated summary

Machine learning for molecular dynamics simulations

This paper evaluates machine learning models for predicting atomic forces and running molecular dynamics simulations. It finds that graph neural networks can accurately reproduce liquid properties when trained on liquid data. But reproducing solid properties requires training on both liquid and solid configurations. Overall, the work underscores the need for comprehensive benchmarking when developing machine learned force fields.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up