Overview

This guide covers the installation of the complete RuNNer suite:

1. RuNNer 2.0 — Ruhr University Neural Network energy representation (Fortran-based)
2. RuNNerASE — The Python interface between RuNNer and ASE (Atomic Simulation Environment)
3. LAMMPS Interface — A RuNNer 2.0 interface to LAMMPS

RuNNer 2.0 Installation¶

We offer two installation routes which trade ease-of-use versus customizability. Choose the one that suits your needs best:

• Conda package: For those looking for a pre-built binary with all dependencies that "just works". No configuration options.
• Build from source: For those looking to customize their build for optimal performance. Developers should choose this option, too.

Additional Components¶

• RuNNerASE: Python/ASE interface for using RuNNer potentials in Python-based simulation workflows.
• LAMMPS Interface: Build LAMMPS with the RuNNer pair style for molecular dynamics simulations.
• Troubleshooting: Common issues and solutions.

Prerequisites¶

Before starting the installation, ensure your system meets these requirements:

For All Parts¶

• Operating System: Linux, macOS, or WSL2 (Windows Subsystem for Linux 2) on Windows
• Git: For cloning repositories
• Build tools: make and/or cmake

For RuNNer 2.0¶

• Fortran Compiler: Intel (ifort/ifx) or GNU (gfortran) supporting Fortran 2008+
• C Compiler: gcc or compatible
• Linear Algebra: Intel MKL (recommended), OpenBLAS, or standard BLAS/LAPACK
• MPI (optional): For parallel training
• FFTW3 (optional): For handling electrostatic interactions in 3G- and 4G-HDNNPs

For RuNNerASE¶

• Python: 3.10 or higher (3.12 recommended)
• uv (recommended): Fast Python package manager
• RuNNer 2.0: Must be built first with library support (make lib) and the RuNNer serial shared library libRuNNer.so must be included in the library path LD_LIBRARY_PATH

For LAMMPS Interface¶

• CMake: 3.10 or higher
• C++ Compiler: Compatible with LAMMPS (Intel or GNU), see LAMMPS documentation for details
• MPI (recommended): For parallel MD simulations using 2Gs and 3Gs. For 4G simulation the serial executable is faster.