Accelerating Metadynamics-Based Free-Energy Calculations with Adaptive Machine Learning Potentials

This repository contains codes for paper Xu, J.; Cao, X.-M.; Hu, P. JCTC, 2021..

This package aims to accelerate metadynamics (MetaD) in heterogeneous reactions using adaptive machine learning potentials (AMLP) that maintains ab initio accuracy.

Installation

Install the external codes, prepare the python environment and add AccMetaD to PYTHONPATH.

Miniconda is recommaned to configure environment.

External Codes

1. VASP 5.4.1
2. DFTB+ 20.1
3. QUIP with GAP

Python Packages

1. ase 3.19.1
2. plumed 2.6.2

Notes

1. Other DFT codes can be utilised as well, which can be accessed by the ase interface.
2. Units in dynamics modules have been changes in ase 3.21.0. Change timestep and temperature accordingly if using new version of ase.

Usage

Introduction

Each job contains at least five input files.

*.xyz is the structure.

plumed-*.dat are inputfiles for plumed.

inputs.py contains DFT, DFTB and GAP calculation parameters.

run.py contains AMLP-MetaD settings.

acc_meta.slurm is the job script that sets environment variables.

Examples

There are four examples attached.

CO on Pt13 cluster using GAP and DFTB-GAP.

CO on Pt(111) surface using GAP and DFTB-GAP.