Freedom251 › Physics Simulation Software
hoomd-tf
What hoomd-tf is
hoomd-tf is a free, MIT-licensed, open-source Linux plugin that connects TensorFlow with HOOMD-blue simulations. It supports online force matching, machine learning during simulations, and collective-variable calculations using TensorFlow tensor operations. GPU workloads can use GPU-to-GPU communication between GPU-accelerated HOOMD-blue and TensorFlow to reduce transfer latency. Built-in utilities include radial distribution function calculation, EDS biasing, and coarse-grained force matching. Users create Keras models as SimModel subclasses, receiving particle neighbor lists, positions, and box data as tensors. MolSimModel can organize positions, neighbor lists, and forces by molecule, including molecules of different sizes. Models may train during a simulation with HOOMD-blue forces as labels, either every step or periodically, and online learning can terminate a molecular-dynamics run early. Documentation covers quickstarts, compilation, tutorials, APIs, and known issues. Python, TensorFlow, and HOOMD-blue are required.
Who it suits
It suits researchers and developers running TensorFlow-based modeling inside HOOMD-blue simulations on Linux, especially GPU users. It is also relevant for molecular simulation workflows needing force matching or collective-variable calculations.
What it does well
- Free and open source under the MIT license
- Supports online force matching and machine learning
- GPU-to-GPU communication targets lower transfer latency
- Includes RDF, EDS, and coarse-grained utilities
- Models can train during simulations
What to know first
- Linux is the only listed platform
- Requires Python, TensorFlow, and HOOMD-blue
- GPU use requires matching CUDA toolkit versions
- Documentation lists a 2021 release
Verdict
hoomd-tf provides a focused TensorFlow integration for HOOMD-blue simulation work, with model interfaces, batching, utilities, and in-simulation training. Setup is best suited to users comfortable managing the listed software and CUDA requirements.
Facts from hoomd-tf's own pages
- Assessment tools
- No · hoomd-tf.readthedocs.io, 23 Sept 2026
- Built-in utilities
- Built-in utilities include RDF calculation, EDS biasing, and coarse-grained simulation force matching. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Early termination
- It supports online machine learning with early termination during molecular dynamics simulations. · hoomd-tf.readthedocs.io, 12 Oct 2026
- GPU execution
- The project is designed to reduce latency with GPU-to-GPU communication between GPU-accelerated HOOMD-blue and TensorFlow. · hoomd-tf.readthedocs.io, 5 Oct 2026
- GPU performance
- The project says it uses GPU-to-GPU communication to minimize latency between GPU-accelerated HOOMD-blue and TensorFlow. · hoomd-tf.readthedocs.io, 12 Oct 2026
- GPU setup limit
- The guide warns that CUDA version mismatches between TensorFlow and HOOMD-blue can complicate GPU setup and recommends compiling HOOMD-blue for GPU use. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Integration
- The documented installation requires Python, TensorFlow, and HOOMD-blue. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Known compatibility constraint
- The guide says TensorFlow and HOOMD must use matching CUDA toolkit versions for GPU use, and recommends compiling HOOMD-blue for GPU mode because of CUDA version mismatches on conda. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Latest documented release
- The changelog lists version 2.4, dated 2021-07-26, with the ability to compute mapped neighbor lists and positions while a simulation runs. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Lesson library
- No · hoomd-tf.readthedocs.io, 23 Sept 2026
- License and cost
- The project paper says HOOMD-TF is freely available under the MIT license on GitHub. · par.nsf.gov, 12 Oct 2026
- Memory control
- For models without molecule batching, tfcompute.attach accepts batch_size to split work into smaller batches for simulations with limited GPU memory. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Model inputs
- The TensorFlow interface exposes HOOMD-blue neighbor lists and particle positions to models. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Model interface
- Users build Keras models as SimModel subclasses, with particle neighbor lists, positions, and box data available as tensors. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Molecule batching
- MolSimModel can arrange positions, neighbor lists, and forces by molecule, including molecules of different sizes. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Optional dependencies
- MDAnalysis and NetworkX are optional; documented utility functions use them for trajectory parsing and coarse-grained graph generation. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Purpose
- HOOMD-TF is a plugin that enables TensorFlow in HOOMD-blue simulations. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Requirements
- The compilation guide lists Python 3.6 or later, TensorFlow 2.3 or later, and HOOMD-blue 2.6.0 or later as requirements. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Support and guidance
- The documentation provides a quickstart, compilation instructions, tutorials, API documentation, and a known-issues section. · hoomd-tf.readthedocs.io, 12 Oct 2026
- Training
- Models can be trained during a simulation using HOOMD-blue forces as labels, with an option to train periodically rather than every step. · hoomd-tf.readthedocs.io, 5 Oct 2026
- Use cases
- It supports online force matching, online machine learning, and collective variable calculations using TensorFlow tensor operations. · hoomd-tf.readthedocs.io, 12 Oct 2026
Where hoomd-tf stands on Freedom251
- Physics Simulation Software#80 of 89