We attempted to repeat the integrations of the other investigators based on the information provided in their publications but were successful only with Rose's orbit (the measure of success was our ability to closely reproduce the published residual statistics). Failure in the other two cases is most likely due to the inability of our software to duplicate the perturbations included by the other investigators. One problem with the Bec-Borsenberger and Rocher integration is that they did not give the value of Titan's mass used in their Titan perturbation.
We were able to match Rose's orbit because he provided state vectors at both
the beginning and end of his integration; we adjusted his initial velocities
slightly to ensure that our integration matched his initial and final
positions. The integration was performed in the FK4/B1950 system with solar
perturbations computed using JPL planetary ephemeris DE102 (Newhall et al.
1995). Rose fit his orbit to 133 observations which he had converted to the
FK4/B1950 system. He included the converted observations in his paper but did
not provide details of his conversion procedure. He found a standard error of
1
52; our reproduction of his orbit gives a standard error of 146 for
those published FK4/B1950 observations. For the pre-1970 observations that we
used our work, the respective standard errors for our orbit and Rose's orbit
are 140 and 178; for the post-1970 observations, they are 060 and
132. The root-mean-square of the differences between our orbit and that
of Rose over the time period 1966-2001 are 4248 km in the radial direction,
10121 km in the in-orbit direction, and 3405 km in the out-of-plane
direction.
Except for the Arequipa data, Bykova and Shikhalev and Bec-Borsenberger and
Rocher fit essentially the same observations. Like Rose, each pair of
investigators converted the observations to the FK4/B1950 system before
processing. Bykova and Shikhalev give a standard error of 15 (with some
unspecified subset of the Arequipa data included), and Bec-Borsenberger and
Rocher quote 
and 
errors of 17 and
12, respectively. For the pre-1982 observations used our work, our orbit
yields respective errors and of
156 and 100.
Bec-Borsenberger and Rocher give sets of polynomial coefficients from which
geocentric astrometric positions can be computed for the years 1981-1990
based on their orbit. From those polynomials we constructed positions for
the first of January of each of those years and compared them to computed
positions from our orbit. In the comparison we assumed that the constructed
positions were in the reference frame of DE102, the planetary ephemeris used
in the analysis, and transformed them with the published rotation between the
J2000 frame and that of DE102.
The rms of and residuals are 026
and 030. It appears that our new orbit agrees with that of
Bec-Borsenberger and Rocher at a level better than our current orbit accuracy.