granthutchison wrote:This is interesting. So the "edge" to the data (apart, that is, from the artefactual 5000 pc sphere) is determined by the fact that there are vanishingly few stars bright enough to allow Hipparcos to detect their parallax at such extreme distances?
Well, yes and no. The average parallax error of the Hipparcos catalogue is about 1 mas, slightly dependent on the apparent magnitude (brighter stars usually have somewhat more accurate parallaxes than fainter ones) but generally independent of the distance to the star.
Thus a measured parallax of (2 +- 1) mas would place the star anywhere between 300 pc and 1000 pc with 1-sigma confidence (68% likelihood), and a parallax of (0.2 +- 1) mas would correspond to a distance between 800 pc and infinity with the same confidence.
The Hipparcos sample is complete down to about V magnitude 8, though some stars in it are as faint as V = 12 mag.
To take this into perspective, the nearest known stellar neigbour of our Sun is Proxima Centauri, at a parallax of 772 mas (distance 1.3 pc) and an apparent V magnitude of 11.0. Thus even out to the very nearest stars the Hipparcos catalogue is not complete down to the luminosity of a Proxima Centauri class star!
At the other extreme we have the red super giants that at a distance of 2000 pc still could be as bright as V = 8 mag.
granthutchison wrote:So could the Hipparcos detectors have detected less than 0.2 milliarcseconds of parallax, if the star had been bright enough?
Hipparcos cannot distinguish parallaxes smaller than 1 mas from zero, even though it could (and did) measure such small parallaxes. That's why one has to take care when using such measurements. The relative
error of the parallaxes just inflates for small parallaxes, as you pointed out yourself.
It is therefore very sensible to clamp distances derived from Hipparcos parallaxes, as done in Celestia. One can be quite certain that all stars seen by Hipparcos must be closer than 5000 pc (except possibly some rare cases). And it really doesn't make much sense using parallaxes that yield infinite relative distance errors anyway. A practical absolute lower limit on sensible parallax measurements would be about 1 mas, corresponding to 1000 pc.
That's all for now, have a very Merry Christmas, all of you!