veldist

Non-parametric Bayesian inference of the line-of-sight velocity distribution (LOSVD) from discrete stellar velocities.

Given individual stellar velocities and per-star measurement uncertainties, veldist recovers the intrinsic LOSVD as a histogram posterior, marginalising over a smoothing hyperparameter that adapts to the signal-to-noise of the data. Designed for resolved stellar kinematics in globular clusters, dwarf galaxies, and the extended halos of nearby galaxies. Includes a batch pipeline for Voronoi-binned data and a writer for the Dynamite histLOSVD / BayesLOSVD input format.

Installation

(pypi coming soon)

pip install -e .

Or for development:

git clone https://github.com/pjs902/veldist.git
cd veldist
pip install -e .

Quick Start

from veldist import KinematicSolver

# Initialize solver
solver = KinematicSolver()

# Set up velocity grid
solver.setup_grid(center=0.0, width=100.0, n_bins=50)

# Add your observational data
solver.add_data(vel=observed_velocities, err=velocity_errors)

# Run inference
samples = solver.run(num_warmup=500, num_samples=1000)

# Plot results
solver.plot_result()

Batch workflow (Voronoi bins) and Dynamite output

Use fit_all_bins to run the full inference pipeline across many bins, then export directly to Dynamite BayesLOSVD files with write_dynamite_kinematics.

from veldist import fit_all_bins, write_dynamite_kinematics

solvers = fit_all_bins(
    bin_data_list,
    grid_kwargs={"center": 0.0, "width": 600.0, "n_bins": 60},
    run_kwargs={"num_warmup": 500, "num_samples": 1000, "gpu": False, "seed": 5567},
    min_stars=10,
)

write_dynamite_kinematics(
    solvers=solvers,
    output_dir="dynamite_input",
    voronoi_bin_metadata=voronoi_bin_metadata,
    bin_flux_mode="nstars",
)

Bins with fewer than min_stars stars are returned as None and written as masked pixels in bins.dat. The Dynamite writer requires astropy. See the Examples page for a complete walkthrough including spatial map extraction.

License

MIT License - see LICENSE file for details.