This is a implementation of High Dimensional Neural Network Potential(HDNNP) designed to reproduce Density Function Theory(DFT) calculation effectively with high flexibility, reactivity.

There is equivalent doc in Japanese README.ja.md.

Install this project by git.

$ git clone https://github.com/ogura-edu/HDNNP.git

# or if using ssh

$ git clone git@github.com:ogura-edu/HDNNP.git

This project uses Pipenv for development workflow. If you don't have it, run this command to install.

macOS

$ brew install pipenv

other

# please run after installing python 
$ pip install pipenv

Same as by anaconda, but you need to install python rather than installing anaconda.

This bug will be fixed in near future release(ref: pythonfinder + pyenv + anaconda issue).

Set environmental variable PIPENV_VENV_IN_PROJECT to 1 to create your VM into this project dir(/path/to/HDNNP/.venv).

export PIPENV_VENV_IN_PROJECT = 1

For macOS users, you need to install mpich before installing dependencies.

# Only for macOS users. 
#
# NOTE: Installing both mpich and openmpi will conflict
#
$ brew install mpich

# or

$ brew install openmpi

Setup your enviroments.

# Install dependencies
$ pipenv install

# activate your VM
$ pipenv shell

# For example...
(HDNNP) $ hdnnpy training

# deactivate
(HDNNP) $ exit

Using anaconda is prefered because it is basically faster than Pipenv.

Install anaconda and activate your VM.

$ ANACONDA_VERSION = [YOUR_ANACODA_VERSION]
$ pyenv install $ANACONDA_VERSION
$ pyenv local $ANACONDA_VERSION
$ conda env create -n HDNNP --file condaenv.yaml
$ echo ". ${HOME}/.pyenv/versions/<anacondaVERSION>/etc/profile.d/conda.sh" > ~/.bashrc

# activate
$ conda activate HDNNP

# install this program using pip
(HDNNP) $ pip install --editable .

# For example...
(HDNNP) $ hdnnpy training

# deactivate
(HDNNP) $ conda deactivate

NOTE

There is no

• ChainerMN
• Chainer v5

on the Anaconda Cloud, so you still have to install these packages by pip.

And these is a bug that if you install anaconda by pyenv, pipenv will fail to start(ref: pythonfinder + pyenv + anaconda issue).

• Jörg Behler. First Principle Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed System, 2007