8.1.2 Astrometric accuracy

The deformation detected versus ICRF2 as as expected, and correct with respect to ICRF3 (Section 8.4.2).

The global zero point has been reduced with respect to Gaia DR2, and now it is estimated to be around $-0.02$ mas (Section 8.4).

Zero-point variations using differential parallaxes of binaries (differential parallaxes of potential common proper pairs) and from open cluster (Section 8.7.1) analysis show a systematic difference between Gaia DR2 and Gaia EDR3.

Tests from Open Clusters show strong variations of the parallax zero point with $G$ magnitude and colour (Section 8.7.1).

Tests from Open Clusters show variations of the proper motion zero point with $G$ magnitude and colour. The effect is more evident in right ascension than in declination (Section 8.7.1).

Tests with APOGEE binaries confirm the large variations of zero-point with magnitude for bright stars, and tests using QSOs also confirm the variation at faint magnitudes (Section 8.4.2).

The large scale regional variations in parallax zero-point are much improved since Gaia DR2 for 5p solutions, both on the average and for fluctuations, but 6p solutions show more pronounced systematics, as shown by the checks using QSOs parallaxes and proper motions (Section 8.2).

Small scale zero-point fluctuations have also been reduced with respect to Gaia DR2, as shown by checks in several regions with different densities (Section 8.2). The effects of the scanning law have been greatly reduced in Gaia EDR3.

In all cases, 6p solutions show a very different behaviour from 5p, and look more prone to systematic effects.

Spurious solutions (objects with very negative parallaxes) have been significantly reduced from Gaia DR2. However, the percentage is still large for 6p solutions and the banding effect (over $\approx 0.6$${}^{\circ }$) is still present.

The zero point also shows variations with (pseudo-)colour for dwarf spheroidals that were not present in Gaia DR2. These variations, however, are not showing in the QSOs.