Introduction to MatPL
Materials Potential Library (MatPL, formerly PWMLFF; current version: MatPL-2026.3) is an open-source software package distributed under the GNU GPL 3.0 license.
MatPL provides a comprehensive collection of software, tools, and data repositories for rapidly developing machine-learning force fields with accuracy comparable to ab initio molecular dynamics (AIMD). The ecosystem includes the MatPL model-training platform, a LAMMPS molecular-dynamics interface, the pwact active-learning data-generation platform, the pwdata format-conversion tool, and repositories for data and models. Use the links below to access their source code and documentation.
Download the MatPL User Manual from Gitee or download it from GitHub
MatPL is also available as a free open-source machine-learning force-field application on the National Supercomputing Internet platform.
1. MatPL Machine-Learning Platform
MatPL rapidly trains machine-learning force fields with accuracy comparable to AIMD. NEP models support efficient large-batch training across multiple GPUs and compute nodes.
This high-performance molecular-dynamics interface integrates MatPL force-field models with GPU acceleration. The NEP interface uses LAMMPS KOKKOS acceleration, provides excellent multi-node parallel efficiency, and supports simulations with more than one hundred million atoms. A single RTX 4090 can handle systems containing more than six million atoms.
PWact is an open-source MatPL-based platform for automated active-learning data generation. It integrates MatPL, the LAMMPS interface, and widely used first-principles packages such as PWmat, VASP, and CP2K. PWact automates job dispatch, monitoring, fault recovery, and result collection, enabling users to generate training datasets that cover broad regions of phase space quickly and cost-effectively.
4. PWdata Structure-Conversion Tool
pwdata is MatPL's data-preprocessing tool for extracting features and labels. It converts structure formats among PWmat, VASP, CP2K, and LAMMPS, and supports supercell construction, lattice scaling, and atomic-position perturbation.
Test results and application examples for MatPL.