Gaia Data Release 2: using Gaia parallaxes

TL;DR

This paper discusses how to accurately infer astrophysical parameters, especially distances, from Gaia DR2 parallaxes using Bayesian methods, addressing challenges like negative and uncertain parallaxes with practical examples and source code.

Contribution

It provides a comprehensive review of traditional methods, introduces Bayesian guidelines for distance estimation from parallaxes, and offers practical tutorials and code for improved analysis.

Findings

Negative parallaxes contain valuable information.

Bayesian methods improve distance estimates.

Practical tutorials facilitate application of methods.

Abstract

The second Gaia data release (GDR2) provides precise five-parameter astrometric data (positions, proper motions and parallaxes) for an unprecedented amount of sources (more than $1.3$ billion, mostly stars). The use of this wealth of astrometric data comes with a specific challenge: how does one properly infer from these data the astrophysical parameters of interest? The main - but not only - focus of this paper is the issue of the estimation of distances from parallaxes, possibly combined with other information. We start with a critical review of the methods traditionally used to obtain distances from parallaxes and their shortcomings. Then we provide guidelines on how to use parallaxes more efficiently to estimate distances by using Bayesian methods. In particular also we show that negative parallaxes, or parallaxes with relatively larger uncertainties still contain valuable information. Finally, we provide examples that show more generally how to use astrometric data for parameter estimation, including the combination of proper motions and parallaxes and the handling of covariances in the uncertainties. The paper contains examples based on simulated Gaia data to illustrate the problems and the solutions proposed. Furthermore, the developments and methods proposed in the paper are linked to a set of tutorials included in the Gaia archive documentation that provide practical examples and a good starting point for the application of the recommendations to actual problems. In all cases the source code for the analysis methods is provided. Our main recommendation is to always treat the derivation of (astro-) physical parameters from astrometric data, in particular when parallaxes are involved, as an inference problem which should preferably be handled with a full Bayesian approach.