Installation

eOn is divided up into two separate programs: a server and a client. The client does most of the computation (e.g. saddle searches, minimizations, and molecular dynamics) while the server creates the input for the client and processes its results.

Getting started

The simplest way to hit the ground running is with the conda package:

# best with pixi
pixi init
pixi add eon
# or with conda/micromamba
micromamba install -c conda-forge eon

At this point any of the many examples should be good to go.

The conda package is a maximalist build with the following potentials and features enabled:

  • Metatomic (machine-learned potentials via libtorch)

  • xTB (semi-empirical tight-binding)

  • Serve mode (rgpot-compatible RPC server)

The server is accessed through python -m eon.server, and the eonclient binary is automatically made available in the activated environment.

Changed in version 2.0: While reading older documentation, calls to eon must now be python -m eon.server.

Obtaining sources

Added in version 2.0: eOn is now developed and distributed primarily via GitHub.

Once git is present[1]:

git clone https://github.com/TheochemUI/eOn.git
cd eOn

Building from source

We provide a conda environment and pixi setup, with dependencies handled by conda-lock.

pixi shell
# or
pixi s -e dev-lite

Other environments can be found by inspecting the pixi.toml file.

This leads to the most robust installation approach:

# conda-compilers may try to install to
# $CONDA_PREFIX/lib/x86_64-linux-gnu
# without --libdir
meson setup bbdir --prefix=$CONDA_PREFIX --libdir=lib --buildtype=release
meson install -C bbdir

Some additional performance can be gained with ccache and mold, which can be passed with --native-file:

  • With ccache installed, add --native-file nativeFiles/ccache_gnu.ini

  • With mold installed, add --native-file nativeFiles/mold.ini

Troubleshooting: rolling distros (Arch, Fedora)

On rolling-release distributions with newer system packages, the conda-forge compiler sysroot can conflict with system headers. The symptom is errors like __iseqsigf128 was not declared or __fpclassify has not been declared when compiling with the pixi/conda compilers.

The root cause is system-installed CMake configs (e.g. nlohmann_json) exporting -I/usr/include, which mixes the system glibc headers with the conda sysroot. Fix by forcing meson to use subproject fallbacks:

meson setup bbdir --prefix=$CONDA_PREFIX --libdir=lib \
  --force-fallback-for=nlohmann_json

Optional packages

The full listing of options is found in the meson_options.txt file. These can all be turned on and off at the command line. As an example see the LAMMPS integration instructions.

For optional wrapped dependencies such as ARTn and IRA, make sure the subproject sources are present before configuring:

meson subprojects download artn-plugin ira

Licenses

eOn is released under the BSD 3-Clause License.

Vendored

Some libraries[2] are distributed along with eOn, namely:

  • mcamc which contains libqd :: BSD-3-Clause license

Added in version 2.0:

  • magic_enum :: MIT License

  • catch2 :: Boost Software License, Version 1.0

  • ApprovalTests.cpp :: Apache 2.0 License

Deprecated since version 2.0:

  • Eigen 2.x :: Mozilla Public License