Hi,
I am working on selecting nuclear candidate transients from the alert stream. I started noticing that there are a couple of objects where the g-band detections are spatially significantly offset from detections in other bands. I attached three examples here:
rubin_offsets.pdf (1017.7 KB)
(The diaObjectId’s in this PDF are links that lead to the corresponding Lasair object page)
In all three cases the detections in other bands are consistent with coming from the center of the host identified in Legacy Survey DR10, while the g-band is off by 0.2 to 0.4 arcseconds. All three objects have many negative subtractions with low reliability score. However, they still contribute to the calculation of the position of the diaObject which ultimately determines the position where the forced photometry is calculated. In the first example, that position is off by 0.2 arcseconds (assuming the host center is the true origin of the detections).
To show that this is a problem in the g-band I attached a histogram showing the offset between the mean position of the individual bands to the mean of all other datapoints (mean position of u-band to mean position of all other bands, etc):
mean_pos_factors.pdf (16.0 KB)
Is this a known issue? If so, should we expect this to continue or is this a problem for the early alerts only? For context, these checks rely on alerts between April 20 and May 7.