plx2dist

Bayesian parallax-to-distance pipeline for stellar catalogues.

Trigonometric parallax is an unbiased distance estimator only at high S/N. At moderate or low S/N — or for the many Gaia stars with negative parallaxes — naive inversion (d = 1/ϖ) fails badly. plx2dist solves this by computing the full posterior p(r | ϖ, σ) on an adaptive grid, combining a Gaussian parallax likelihood with a physically motivated distance prior (EDSD or uniform volume density). It was developed for CNS6 (The Sixth Catalogue of Nearby Stars) but is general-purpose for any Gaia-era parallax catalogue.

Installation

pip install plx2dist

Quick start (Python API)

import pandas as pd
from plx2dist import derive_distances

df = pd.read_csv("my_catalogue.csv")   # must contain 'parallax', 'parallax_error'

if __name__ == "__main__":
    result_df = derive_distances(df, plx_col="parallax", err_col="parallax_error", n_jobs=4)

Quick start (command line)

plx2dist --input catalogue.vot \
         --output-prefix results \
         --priors edsd \
         --n-jobs 8

Output columns

For each prior (e.g. edsd), the following columns are added:

Column	Description
distance_edsd_mode	Posterior mode (pc)
distance_edsd_q05	5th percentile (pc)
distance_edsd_q16	16th percentile — lower 1σ bound (pc)
distance_edsd_q50	Median distance (pc) — recommended point estimate
distance_edsd_q84	84th percentile — upper 1σ bound (pc)
distance_edsd_q95	95th percentile (pc)
distance_edsd_err_lo	q50 − q16 (pc)
distance_edsd_err_hi	q84 − q50 (pc)
distance_edsd_p_within_25pc	P(r < 25 pc | ϖ, σ) — membership probability; threshold can be specified by user
distance_edsd_r_grid_min_used	Diagnostic: refined grid lower bound (pc)
distance_edsd_r_grid_max_used	Diagnostic: refined grid upper bound (pc)
distance_edsd_prior_L_pc	EDSD scale length used (pc)

Use q50 as the point estimate and err_lo/err_hi as the asymmetric uncertainty. mode tends to be slightly closer to the likelihood peak; prefer q50 for catalogue work.

Method

A fine grid from 0.1 to 10 kpc turned out to be too slow and missed low-S/N posteriors near the boundary.

The pipeline ended up doing two passes - adaptive log-r grid:

1. a coarse safety pass to locate where the posterior actually lives.
2. then a refined pass around that region for accurate quantile estimation.

No MCMC. Fully deterministic and parallelised via Python multiprocessing.

Priors

Name	Formula	Default L
edsd	p(r) ∝ r² exp(−r/L)	250 pc (tuned for CNS6 volume)
volume	p(r) ∝ r² (uniform space density, truncated)	—

For the solar neighbourhood (e.g., ≲ 25 pc) the EDSD prior with L = 250 pc is recommended.

Reference

If you use plx2dist, please cite:

• Golovin et al. (in prep.) — CNS6 catalogue and distance methodology

BibTeX for CNS6:

@article{Golovin2026,
  author  = {Golovin, A. et al.},
  title   = {{The Sixth Catalogue of Nearby Stars (CNS6)}},
  journal = {A\&A},
  year    = {2026},
  volume  = {NNN},
  pages   = {ANN},
  doi     = {TBD}
}