mff 1.1.2

Gaussian process regression to extract non-parametric 2-, 3- and many-body force fields.

Machine learning nonparametric force fields (MFF)

An example tutorial jupyter notebook can be found in the tutorials folder.

Table of Contents

• Background on MFFs
• Install
• Examples
• Maintainers
• References

Background on MFF

The MFF package uses Gaussian process regression to extract non-parametric 2- and 3- body force fields from ab-initio calculations. For a detailed description of the theory behind Gaussian process regression to predict forces and/or energies, and an explanation of the mapping technique used, please refer to [1].

For an example use of the MFF package to build 3-body force fields for Ni nanoclusters, please see [2].

Pip Installation

To install MFF with pip, simply run the following in a Python 3.6 or 3.7 environment:

pip install mff

Source Installation

If the pip installation fails, try the following: Clone from source and enter the folder:

git clone https://github.com/kcl-tscm/mff.git
cd mff

If you don't have it, install virtualenv:

pip install virtualenv	   

Create a virtual environment using a python 3.6 installation

virtualenv --python=/usr/bin/python3.6 <path/to/new/virtualenv/>	

Activate the new virtual environment

source <path/to/new/virtualenv/bin/activate>	

To install from source run the following command:

python setup.py install	

Or, to build in place for development, run:

python setup.py develop

Examples

Refer to the two files in the Tutorial folder for working jupyter notebooks showing most of the functionalities of this package.

Maintainers

• Claudio Zeni ([e-mail]),
• Aldo Glielmo ([e-mail]),
• Ádám Fekete ([e-mail]).

References

[1] A. Glielmo, C. Zeni, A. De Vita, Efficient non-parametric n-body force fields from machine learning (https://arxiv.org/abs/1801.04823)

[2] C .Zeni, K. Rossi, A. Glielmo, A. Fekete, N. Gaston, F. Baletto, A. De Vita Building machine learning force fields for nanoclusters (https://arxiv.org/abs/1802.01417)