Consider statistical implications of special situations

[taldcroft]

The Orion field is an interesting case. It has many identified sources (though I'm not sure how many are in Tycho). From past discussions there is at least one Tycho match that was observed 22 times and is a known double. So you have two things happening:

1. One oft-observed source with a systematically bad counterpart putting a lot of bad samples into the dataset.
2. One observation with a lot of sources putting a lot of repeat values of offset into the dataset.

Issue 1 can presumably be addressed by making a table of bad source positions, so X-ray sources within 5 arcsec of a bad source position is not used in statistics or most plots (or gets marked as bad in any visualizations).

Issue 2 may come down to generating a per-obsid offset that represents the weighted average for that obsid. We might have talked about that before. This does open a new source of problems, say you have 4 X-ray sources in a field and 3 of them have small radial offsets < 0.5 arcsec and one is an outlier at 2.0 arcsec, but there is no obvious justification for throwing that one out.

Old Orion discussion

TA "Yes, this works best for named sources. Orion nebula has hundreds of sources (https://www.nasa.gov/mission_pages/chandra/images/the-orion-nebula.html) with good astrometry but again it is possible that one could be dealing with a binary. A positional search is reasonably likely to give the right source most of the time in this case.

JC In the Orion nebula observations, it looks like what we're probably matching is this double star: http://simbad.u-strasbg.fr/simbad/sim-id?Ident=%40810775&Name=*%20tet01%20Ori%20C&submit=submit

TA

This does point out that the single star in Orion was apparently contributing 22 of the 107 data points, which clearly indicates a need to control for repeats. This is not entirely trivial because it is difficult to separate out systematic offsets that are real (due to the X-ray emission coming from a different place than the identified optical source in our catalog) vs. systematic celestial location errors (i.e. all using the same fid set) vs. statistical cel. loc. errors. I don't have any immediately great answer to that.

cc: @javierggt @jeanconn

[javierggt]

I don't remember discussing this, so I guess it was before time (?). In any case it is interesting, I did see a couple of these OBSIDs in this iteration and only one source had a counterpart, which was a bit disappointing.

I think handling repeats is interesting. In this case, I suppose proper motions will be significant and correlated, so one cannot just take two fields at different times and apply a transform.

[javierggt]

22334 22335 22336 22337 22338 22340 22341 22342 22343 22892 22893 22904 22993 22994 22996 22997 22998 22999 23000 23001 23002 23003 23004 23005 23007 23008 23010 23011 23012 23087 23097 23114 23115 23120 23206 23207 23208 23233 24622 24623 24624 24829 24830 24831 24832 24834 24842 24873 24874 24906

[taldcroft]

This highlights the need for attention to the counterpart source catalogs. For instance in Orion one can get (from a paper) coordinates for all the X-ray/optical counterparts. They are also likely all in Gaia, but the issue there being that Gaia is huge so processing is slower and there will be more false positives.