Hi John,

Regarding that five-minute run with the entire dataset: you
should probably set up Find_Orb to make use of JPL DE ephemerides.
Doing that should bring you down to the ballpark of ten seconds.
This is described a bit at

http://www.projectpluto.com/find_orb.htm#de_eph

Use of the pre-1990 observations will not (in general) be
very useful, simply because their accuracy usually isn't as
good as more recent astrometry using modern catalogs and CCDs.

Find_Orb doesn't do any automated selection of observations.
(At least, not yet.) For this object, all I did was to select
the observations from the three observatories mentioned in my
previous post: (673) Table Mountain, (689) Flagstaff, and
(950) La Palma. These observatories all are deliberately going
after highly precise astrometry, with observations reduced using
Tycho-2 or UCAC2. There are probably some others (for example,
Gordon Garradd at (422) Loomberah has gotten some excellent results),
but not many, and I don't know who they might be. Perhaps Jan
Manek or Rob Robinson or Steve Preston could comment?

As you noted, if you click on Herget, Find_Orb doesn't use
all the data; it just selects a subsection containing a few
hundred days' worth of data. That's because the method of Herget
doesn't work well on longer arcs (details are given on the Find_Orb
page,

http://www.projectpluto.com/find_orb.htm

The methods of Vaisala and Gauss are similarly limited. All of
these are really intended just to get an initial orbit (and in the
case of the method of Herget, to refine that orbit a little). Once
you have a good initial solution, the "full step" (least squares)
method can be used, without running into such limitations... a
full explanation is quite lengthy; it's given at the above page,
and truly gung-ho orbit computers would probably say I've barely
scratched the surface of this topic.

-- Bill