brucefan1983/GPUMD

What it solves

GPUMD (Graphics Processing Units Molecular Dynamics) provides a high-performance framework for running large-scale atomistic simulations on GPUs. It specifically addresses the computational bottleneck of traditional molecular dynamics by enabling the use of Neuroevolution Potentials (NEPs), which combine the high accuracy of quantum mechanical calculations with the speed of empirical potentials.

How it works

The project consists of two main programs: gpumd for performing molecular dynamics simulations and static calculations, and nep for training and predicting with Neuroevolution Potentials. Users define their system via input files (run.in, model.xyz) and leverage GPU acceleration to simulate complex physical phenomena like heat transport, mechanical properties, and shock compression. The framework supports both empirical interatomic potentials and machine-learned NEP models.

Who it’s for

This tool is designed for computational materials scientists, physicists, and researchers working on atomistic simulations who need to perform large-scale calculations with high efficiency. It is suitable for those looking to integrate machine-learned potentials into their research workflows.

Highlights

• High-performance GPU acceleration for molecular dynamics simulations.
• Built-in support for Neuroevolution Potentials (NEPs) for accurate, efficient modeling.
• Versatile application range including heat transport, thermodynamic properties, and shock compression.
• Extensive documentation, tutorials, and Google Colab integration for rapid onboarding.
• Tools provided for data processing, format conversion, and structure preparation.