Skip to content

torch-admp

Automatic Differentiable Multipolar Polarizable (ADMP) in PyTorch backend

Overview

torch-admp is a PyTorch implementation of the ADMP (Automatic Differentiable Multipolar Polarizable) module in DMFF (Differentiable Molecular Force Field) package. This package provides efficient implementations of various molecular dynamics force calculations including:

  • Particle Mesh Ewald (PME) for electrostatic interactions
  • Charge Equilibration (QEq) methods
  • Polarizable electrode models
  • Neighbor list management
  • Optimization algorithms

Installation

pip install torch-admp

For development:

git clone https://github.com/ChiahsinChu/torch-admp.git
cd torch-admp
pip install -e .[docs,test]

Features

  • GPU Accelerated: Built on PyTorch for efficient GPU computation
  • JIT Compilation: Support for TorchScript compilation
  • Modular Design: Clean separation of different force components
  • Extensible: Easy to add new force modules

Documentation

Citation

If you use torch-admp in your research, please cite:

@software{torch_admp,
  author = {ChiahsinChu},
  title = {torch-admp: ADMP in PyTorch backend},
  url = {https://github.com/ChiahsinChu/torch-admp},
  year = {2024}
}

License

This project is licensed under the LGPL-3.0-or-later License - see the LICENSE file for details.