skip to main content

gaia data release 3 documentation

11.4 Quality assessment and validation

11.4.3 Distances

Author(s): Rene Andrae, Coryn A.L. Bailer-Jones, Orlagh L. Creevey, Morgan Fouesneau, Antonella Vallenari

Table 11.38: Distance estimates in Gaia DR3.
distance GSP-Phot
Figure 11.77: Comparison of the GSP-Phot distances to literature values for clusters from Cantat-Gaudin et al. (2020). Each errorbar indicates the median and [16th, 84th] percentile of all the members of a cluster. For single stars in clusters, GSP-Phot tend to systematically underestimate distances above 2 kpc depending on fractional parallax uncertainty, σϖ/ϖ.
Figure 11.78: Comparison of GSP-Phot distances with parallaxes for a random subset of one million stars. Panel (a): Parallax vs. distance, colour-coded by parallax signal-to-noise ratio ϖ/σϖ. The black dashed line indicates the expected relation ϖ=1d in absence of noise. Black contours indicate density dropping by factors of 3. Panel (b): Distributions of distances (red) and inverse parallax (black) for all sources in this random subset. Panel (c) and (d) are identical to panel (b) but only for sources with ϖ/σϖ>5 and 10, respectively. We caution that the published parallaxes do not contain the zeropoint correction used to obtained the distances.

Two Apsis modules provide us with distance estimates: GSP-Phot (see Section 11.3.3) for single stars and MSC (see  Section 11.3.5) for unresolved binary systems.

Single Stars

For single stars, GSP-Phot provides reliable distances out to 2kpc. Beyond this limit, GSP-Phot can systematically underestimate distance depending on the fractional parallax uncertainty, σϖ/ϖ. In Figure 11.77, we compare the distances from GSP-Phot with literature values for each cluster analysed in Cantat-Gaudin et al. (2020). This catalogue used the Gaia DR2 data and a maximum likelihood approach, but we updated their results on parallaxes for the DR3 zero point. We emphasize that our comparison does not depend on Gaia DR2 systematics on astrometry. We obtain similar results when comparing GSP-Phot distances with those derived from color-magnitude diagram fitting by Kharchenko et al. (2013) and BOCCE (Bragaglia and Tosi 2006; Cantat-Gaudin et al. 2018).

However, when the fractional parallax uncertainties are small (about σϖ/ϖ<0.2 or ϖ/σϖ>5), the distances from GSP-Phot remain overall reliable up to 10 kpc (see Figure 11.78). Yet, Figure 11.78b-d shows that the distributions of GSP-Phot distances and inverse parallaxes can be markedly different, depending on the parallax quality. We emphasize that the large discrepancy between GSP-Phot distances and the inverse parallaxes in Figure 11.78b is not fully explained by “inverting a noisy parallax measurement”. There is a real systematic underestimation of GSP-Phot distances that is most likely due to the GSP-Phot distance prior. When the parallax measurements are of sufficiently high quality (small uncertainties), the prior has little to no effects, which restores the agreement between the distances and parallaxes as shown in Figure 11.78d.

Figure 11.79: Comparison of distances obtained from GSP-Phot from Gaia DR3 (blue histogram), when using a more relaxed distance prior (cyan histogram) and when using no priors at all (red histogram). We compare to distributions of inverse parallax from Gaia DR3 (black histogram) and inverse parallaxes with zero-point correction as used during CU8 processing (grey histogram).

To investigate that the GSP-Phot distance prior is too harsh, we have reprocessed 5 million sources with two additional different priors definitions. The distance estimates published in Gaia DR3, used an exponential distance prior with a scale length of 1/10 of the one derived from the mock catalogue of Rybizki et al. (2020) (see Section 11.3.3). In one reprocessing, we restored the actual length scale from the mock catalogue of Rybizki et al. (2020) (i.e., increased the Gaia DR3 one by a factor of 10.). In another reprocessing, we remove all priors, i.e., uniform up to 10 kpc. Figure 11.79 shows the resulting distance distributions. A weaker prior would have made GSP-Phot distances more consistent with the parallaxes, thereby confirming that the root cause of the distance bias issue. However, Figure 11.79 also shows that a the uniform prior is also non-optimal.

Figure 11.80: Comparison of distances with asteroseismically-derived distances for main sequence stars from Huber et al. (2017) (left) and for giants from Anders et al. (2017) (right), colour-coded by parallax_over_error. The bisector is indicated in red.

We compared GSP-Phot distances with those from asteroseismic analysis in Figure 11.80. Overall, we find a good agreement out to 2–3 kpc for both main-sequence and giant stars. Beyond this limit, we find more outliers and observe a systematic underestimation of distance, particularly for lower SNR on the parallax.

Unresolved Binary Systems

Figure 11.81: Comparison of distances to parallaxes for MSC (panel a) and GSP-Phot (panel b) for spectroscopic binaries from Pourbaix et al. (2004) (red points) and Traven et al. (2020b) (blue points). Both panels show the exact same set of stars.

We combined the spectroscopic binaries from Pourbaix et al. (2004) and Traven et al. (2020b) to obtain a reference sample to investigate the quality of MSC’s distances. On this sample, we find that the distances from MSC are largely consistent with the parallaxes (see Figure 11.81a). Surprisingly though, Figure 11.81b shows that for these known binaries GSP-Phot distances that assume single stars agree better with the Gaia parallaxes than the MSC distances. We suspect that the spectroscopic binaries in Pourbaix et al. (2004) and Traven et al. (2020b) are so bright that their excellent parallaxes (small uncertainties) override the single-star assumption by GSP-Phot. MSC does not use the parallaxes in the same manner as GSP-Phot. A sample of fainter spectroscopic binaries should ideally show that MSC distances agree better with parallaxes than GSP-Phot distances.