# 10.2.2 Duplicated sources

The detection system on-board Gaia may sometimes generate more than one detection for a given source. The on-ground cross match process will then have to decide if two simultaneous detections represent two sources or merely repeated detections of the same single source. With dozens of thousands of millions detections, chances are that some small fraction of the single sources end up with part of their observations labelled with one source identifier, and the rest with a different source identifier. When this happens, the various pipelines produce two sets of independent results each with their own source identifier. We will of course expect, that the two sets are essentially identical. In the final catalogue only one solution is kept and the other removed, and the published source carries a flag saying it was a ‘duplicated source’. In deciding which solution to keep, priority is given to the one with better astrometric results. Before removing the poorer solution, this set of duplicated sources from the ‘pre-DR2’ Catalogue provided an attractive opportunity for the catalogue validation. Some results are presented in Arenou et al. (2018) and others in Section 10.2.5, typically pointing to a certain underestimation of the errors.

First, however, we must decide when to consider that a close source-pair is a duplicate. From simple considerations on the image size, data acquisition, and the present state of the on-ground processing, sources can hardly be expected to be reliably resolved if they are separated by less than 0.3 ${}^{\prime\prime}$, and even that is an optimistic limit as long as no dedicated data processing for binaries is activated.

Figure 10.6 shows the relation between separation and magnitude difference for sources in a dense test field in a processing step immediately before duplicates were removed. The sign of the magnitude difference depends, for processing reasons, on which source had the highest declination and can be considered random. We notice several characteristics: a concentration at small separations (below 0.05 ${}^{\prime\prime}$) and magnitude differences; a continuation up to 0.4 ${}^{\prime\prime}$ separation with small magnitude differences; a widening up to 0.7 ${}^{\prime\prime}$; and a more gradual widening at higher separations. A tentative interpretation is that the concentration at very small separations then represents the well-behaved, single sources. The, relatively sparse, continuation to 0.4 ${}^{\prime\prime}$ is a mixture of genuine duplicate sources and actual binaries; the widening up to 0.7 ${}^{\prime\prime}$ show the gradual decrease of acquisition conflicts; and the final widening demonstrates the continued decrease of conflicts and the increasing ability to deal with magnitude contrasts. Typical acquisition windows are 0.7 ${}^{\prime\prime}$ by 2.1 ${}^{\prime\prime}$, so two sources closer than 0.7 ${}^{\prime\prime}$ will always be in conflict.

Minor conflicts, affecting a small part of the acquisition windows, are accepted in Gaia DR2, and for pairs with similar magnitudes either of them may be detected as the brighter in a given transit and get a complete window. Even so, the closeness of a neighbour of similar magnitude will be disturbing. All in all it was decided to define duplicate sources for Gaia DR2 as pairs closer than 0.4 ${}^{\prime\prime}$ at the reference epoch. The vast majority of these pairs are genuine duplicates and only a minor fraction are binaries.

Figure 10.7 shows the separations for source pairs identified as duplicates. Because of the way duplicates involving more than two sources are handled, separations may occasionally exceed 0.4 ${}^{\prime\prime}$. The left panel shows a strong peak in the centre with a short tail that gradually becomes more rectangular and end in a hint of a diagonal cross. As shown in the right panel, the more exotic behaviour only occurs for cases where the full astrometric solution had to be abandoned.