Introduction

HAL’s MD package is a high-precision molecular dynamics framework for the simulation of complex dynamic phenomena in materials research, in a broad sense, with a focus on nanofluidics and soft matter. The software has been specifically designed for acceleration through CUDA-enabled graphics processors, with an emphasis on both high performance and numerical long-time stability. It shows very good scalability from few thousands to millions of particles.

HAL’s MD package is maintained and developed by Felix Höfling and was initially written together with Peter Colberg. Special credit goes to Nicolas Höft and Daniel Kirchner for their manifold contributions.

Note

A description of the implementation, performance tests, numerical stability tests, and an application to slow glassy dynamics can be found in the article by P. H. Colberg and F. Höfling, Highly accelerated simulations of glassy dynamics using GPUs: Caveats on limited floating-point precision, Comput. Phys. Commun. 182, 1120 (2011) [arXiv:0912.3824].

Physics applications

HAL’s MD package was made to study

  • the spatio-temporal dynamics of inhomogeneous and complex liquids
  • glass transition, liquid–vapour interfaces, phase separation of binary fluids, confined fluids, porous media, … (active matter coming soon)
  • particles interacting via many truncated and untruncated potentials of the external and pair type (anisotropic and bonded potentials coming soon)
  • microcanonical (NVE) and canonical (NVT) ensembles with Hamiltonian and Brownian dynamics (Integrators, optimised multi-stage integrators coming soon)
  • both two- and three-dimensional systems, natively
  • with periodic, closed and open boundaries

Features

HAL’s MD package features

  • GPU acceleration: 1 NVIDIA A40 GPU shows 200-300 fold speed-up over a single Xeon CPU core (Benchmarks)
  • high performance and excellent numerical long-time stability (e.g., energy conservation)
  • user scripts, which define complex simulation protocols
  • online evaluation of observables including dynamic correlation functions
  • structured, compressed, and portable H5MD output files
  • extensibility by the generic and modular design
  • free software under LGPL-3+ license

Technical features

HAL’s MD package brings

  • an extensive automatic test suite using CMake
  • double-single floating-point precision for numerically critical hot spots
  • C²-smooth potential cutoffs for improved energy conservation
  • an integrated, lightweight Lua interpreter
  • modern, template-based C++ code taking advantage of C++14

Historical footnote

The name HAL’s MD package was chosen in honour of the machine HAL at the Arnold Sommerfeld Center for Theoretical Physics of the Ludwigs-Maximilians-Universität München. HAL has been the project’s first GPGPU machine in 2007, equipped initially with two NVIDIA GeForce 8800 Ultra. HAL survived a critical air condition failure in the server room.