Hi Luc,

> I mentioned this problem last week to Jon Giorgini from the JPL...

I did think it was probably not a coincidence!

> If you look at the algorithm of NG Roman, you will notice that the
> boundaries converted in decimals are rounded to 4 decimals.

Hmmm... hadn't realized that. I think it must have been extended
to five places before I ever saw the data, reducing the maximum error
to 1e-5 hours in RA, or .54 arcseconds. I've been using the versions at

ftp://cdsarc.u-strasbg.fr/pub/cats/VI/49/

(which you will see includes my somewhat modified version, with a
few small corrections and constellations specified for _both_ sides of
each segment, plus some documentation. And I think I will submit a
revision of my file to CdS with extra digits in RA and dec.)

I notice that the 'bound_20.dat' file on the CdS site uses seven
places. However, if it was generated by reading in 'bound_18.dat'
with its five-place data, precessing to J2000, and then outputting
the results to seven places... then we have extra digits, but not
actual extra precision.

> 1) The precession formulas are old formulas from Newcomb, I am not sure
> that they may still be used with J2000 positions.

I'd think there would at least be no harm in using more modern formulae.
(I don't actually know how much the Newcomb formulae differ from current ones.
I doubt it would account for the difference of about 18" we're seeing for
the case you mention, though.)

> 2) It looks like the time is badly rounded for T and t in the precession
> formulas like just entering 2000 and 1875 for the calculation rather
> than the accurate Julian day time 2451545 for J2000 and 2405889.25855
> for B1875.

I doubt it would make a huge difference -- certainly not an 18" one --
but again, it shouldn't hurt to do the math accurately.

> And after all that there is the aberration question.

True. And the difference you're seeing is an aberration-sized one.

But again, I am reasonably confident that the boundaries ought to
be defined to be "fixed" in either B1875 or J2000 coordinates; i.e., a
star with near-zero proper motion and parallax shouldn't change constellations
twice each year, simply because its apparent coordinates cross a particular
RA or dec boundary. (Nor twice every eighteen years as the earth nutates.)

-- Bill