# 8.1.3 Astrometric precision

• The formal uncertainties have much improved since Gaia DR2.

• The quasi-gaussianity of the uncertainities, checked using QSOs, still holds. The normality is better, as expected, for 5p than for 6p solutions the latest contain a significant number of non-single stars.

• Parallax uncertainties seem to be underestimated, as shown by deconvolution versus their formal uncertainties, for various subsets and also by using LMC subsets (Section 8.2). The underestimation is larger for brighter magnitudes, and generally larger for 6p solutions .

• For 5p solutions the underestimation reaches 1.35 for the most precise parallaxes, decreasing and reaching a flat curve for the less precise ones. For this later case the total uncertainty $\sigma_{ext}$ is well described by the expression $\sigma_{ext}^{2}=k^{2}\sigma_{int}^{2}+\sigma_{sys}^{2}$, where $\sigma_{int}$ is the formal error given in the catalogue and $\sigma_{sys}\approx 15\mu as$. See Gaia DR2 known issues page, document on astrometry.

• Comparison of proper motions with proper motions estimations obtained from Hipparcos and Gaia position show a strong variance factor variation with magnitude and colour, which was not as strong for Gaia DR2. Checks using QSOs show the same behaviour.

• Astrometry is apparently ‘too good’ for bright stars in crowded areas (e.g. in Baade’s window), meaning too good astrometric_gof_al, astrometric_excess_noise and ruwe. It might be that some extra noise was absorbed by the excess attitude noise (Section 8.2).

• Proper motion uncertainties in crowded fields estimated using globular clusters (Section 8.7) are correctly estimated for stars having little or no contamination. Stars with large contamination are on average 1.6-1.8 sigma from the expected value indicating that uncertainties on proper motions are underestimated. This shows a substantial improvement in comparison with Gaia DR2.

• Unresolved binaries in open clusters are not identified by any of the parameters: astrometric_excess_noise, astrometric_gof_al, ipd_frac_multi_peak, ruwe, ipd_gof_harmonic_phase, ipd_gof_harmonic_amplitude (Section 8.7).