cuRDF 0.4.0

GPU-accelerated radial distribution functions with NVIDIA ALCHEMI Toolkit-Ops neighbor lists and PyTorch.

cuRDF

CUDA-accelerated radial distribution functions using NVIDIA ALCHEMI Toolkit-Ops O(N) neighbor lists and PyTorch. Compatible with ASE Atoms or MDAnalysis Universe objects.

Benchmarking

cuRDF is benchmarked against RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.

Install

Latest release:

pip install cuRDF

For development:

git clone https://github.com/joehart2001/curdf.git
cd curdf
pip install -e .

Quickstart

ASE Atoms object:

from ase.io import read
from curdf import rdf

# Load trajectory or frame e.g. XYZ, extxyz, traj, LAMMPS data/dump
atoms = read("md_run.extxyz")

# Compute RDF between species C and O from 1.0 to 8.0 Å
bins, gr = rdf(
  atoms,
  species_a="C",
  species_b="O", # species b can be the same as species a
  r_min=1.0,
  r_max=8.0,
  nbins=200, # resolution of rdf histogram binning
  method="cell_list", # neighbor list method: "cell_list" (larger systems) or "naive" (smaller systems, less overhead)
  output = "results/rdf.csv" # optional output
)

# Plot RDF
plot_rdf(bins, gr, path="results/rdf.png")

MDAnalysis Universe (topology and trajectory):

import MDAnalysis as mda
from curdf import rdf

u = mda.Universe("topology.data", "traj.dcd", atom_style="id type x y z")
bins, gr = rdf(
  u,
  species_a="C",
  species_b="O",
  r_min=1.0,
  r_max=8.0,
)

Citation

If you use cuRDF in your work, please cite:

@software{cuRDF,
  author    = {Hart, Joseph},
  title     = {cuRDF: GPU-accelerated radial distribution functions},
  month     = dec,
  year      = 2025,
  publisher = {Zenodo},
  version   = {0.1.0},
  doi       = {10.5281/zenodo.1085332119},
  url       = {https://doi.org/10.5281/zenodo.1085332119}
}