The WCS information written at the telescope is only approximate.

It has the following characteristics:
<LI> Tangent plane projection.  This projection does not properly account for the distortion introduced by the refractive corrector.
<LI> RA and Dec center values are read from the telescope.  This may be in error by an arcsecond or two if the collimation of the telescope drifts or was not initialized properly.
<LI> Position angle.  The rotator angle is observed to have a 0.5 degree error relative to what is expected nominally.  The actual position angle is 0.5 less than what is written in the header.
<LI> Chip locations.  The header contains the nominal locations of all the CCDs in the mosaic.  In practice each chip is a few pixels off from the nominal location.
<LI> Nominal pixel scale.  Actual pixel scale will be affected by refraction effects.

<P>

The initial WCS is adequate for object identification, but not for precision astrometry.

<P>
The final WCS has the following characteristics

<LI> Uses the ZPN projection described in Calabretta et al XXX.  This
     is a projection uses a polynomial to convert from angle on the
     sky to position on the focal plane.  The polynomial coefficients
     were derived from the optical design of the MMT Wide Field
     corrector.  We have seen no evidence that this representation is
     insufficient.  The downside to the ZPN projection is that it is
     not understood by IRAF.  However, swarp and sextractor both
     understand the ZPN projection (after patching).  If anyone knows of a polynomial
     representation that is understood both by ds9 and by IRAF, please
     let me know.

<LI> The chip positions are set to a set of pre-determined positions.

<LI> All 6 members of the CD matrix are solved for, namely the image
     center, rotation, pixel scale and shear terms.  This solution is determined
     globally for the entire mosaic  We assume that the
     geometry of the focal plane is stable and thus we don't need a
     large number of stars on each chip.  In practice we achieve fits to the 2Mass catalog
     with 0.13" RMS -- at the level of the catalog.

<P>

The nuts and bolts of WCS determination.

The wcs determination has 3 steps:

<LI> 1.  Run the IRAF script fixmosaic.  This corrects the rotator angle, and if necessary any 
    gross error in the image center.
<LI> 2.  Run the IRAF script zpn.  This changes the CTYPEn header keywords from TAN to ZPN to complete
    the WCS installation.
<LI> 3.  Run the megawcs script.  This script runs SExtractor, matches the stars to a catalog, and
    then determines WCS parameters and writes them into the header.


megawcs details