Supplementary information for the Pull Request #4898 in CP2K
Abstract
This repository provides supplementary material for Pull Request #4898 to CP2K. It contains the input files, trained models, datasets, and benchmarking scripts used to validate the updated implementation of NequIP and Allegro in CP2K against their reference implementations in LAMMPS. Both Allegro and NequIP models were trained on the same dataset consisting of 32 water molecules generated from AIMD simulations at different temperatures using the SCAN functional. The following software versions were used: Allegro 0.8.0 NequIP 0.16.0 Contents of the archive dataset/ water_all_stress_evA3.extxyzAssembled dataset including: potential energies forces stress tensor (in eV/ų) cell information atomic_energies/Contains the isolated atomic reference energies for oxygen and hydrogen. allegro/ Training input files Compiled Allegro model parity/ directory containing: lammps/ – MD trajectory generated with LAMMPS cp2k/ – CP2K reference trajectory evaluated on the same LAMMPS trajectory nequip/ Training input files Compiled NequIP model parity/ directory containing: lammps/ – MD trajectory generated with LAMMPS cp2k/ – CP2K reference trajectory evaluated on the same LAMMPS trajectory Benchmarking scripts cp2k-vs-lammps.ipynbJupyter notebook used to generate parity plots comparing CP2K and LAMMPS energies, forces, and stress tensors. plotting_cp2k_vs_lammps.pyPython script used to parse output files from LAMMPS and CP2K and generate comparison plots.
Fields of Study & Subjects
Machine Learning in Materials Science
Advanced Chemical Physics Studies
Advanced Physical and Chemical Molecular Interactions
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 3 March 2026
- Language
- EN