Thanks for the clarification. Jamie was coming to this conclusion
yesterday afternoon but failed to respond because we were all off to see
Return of the King! I think we are edging towards wanting GAIA to default
to the FITS standard.

Tim

Here is Jamie's response:

Date: Thu, 18 Dec 2003 09:40:11 -1000 (HST)
From: Jamie Leech <[log in to unmask]>
To: Tim Jenness <[log in to unmask]>
Subject: Re: Projections in AST (fwd)


> As you thought. Can you form a response please? Also, note that your LSR
> header is incorrect.

 The short answer is that David Berry has just said what was I trying to
describe to you yesterday on the white board, and I agree completely.
Essentially CAR and MER are only rectangular in the native (or
intermediate system) not celestial co-ordinates. If the celestial
co-ordinate of the ref pixel happens to be RA=0 Dec=0, the grid lines will
look rectangular, otherwise they won't.   It looks as if

1) GAIA does not follow the FITS-WCS paper III convention by default (goes
to the CARLIN thing). GAIA can, and probably should, follow FITS WCS by
default, or at least give an astrometry warning when reading a fits file
and deciding to do something different. The CARLIN mode rectangular grid
should be plotted over the entire map still, with dec increasing to the 90
line then decrerasing again after the 90 line, with a 180 degree flip in
RA.

2) The fv fits viewer does not do the correct FITS WCS paper II thing
when presented with CAR and MER co-ordinates.

3) AIPS++ seems to do the right thing.

 It is clear that AIPS++ image to FITS writing may still need work when it
comes to writing out the spectral part of the header correctly, however.

 Jamie.


On Thu, 18 Dec 2003, David Berry wrote:

> Tim,
>     Sorry for the delay, I was off work yesterday.
>
>
> > All have [1,1,1] set at 05:20:04.000 +89.59.00.000 6.914730e+11 Hz
> > [The spectral headers are not liked by AST v3 so I need to look into
> > that].
>
> The problem is the "-LSR" in the CTYPE3 value. The final 4 characters
> should indicate the algorithm used for projecting frequency onto pixel
> number. I guess your data is linear, so CTYPE3 should be just "FREQ".
> The standard of rest should go in the SPECSYS keyword which should be set
> to "LSRK" or "LSRD".
>
>
> > You'll be pleased to hear that GAIA does correctly report the
> > RA/Dec for CAR as the pole is crossed but the axis overlay is not quite
> > correct since it stops at +90 and does not draw in the top half of the
> > grid. I assume this is because of the RA discontinuity. Any chance that
> > could be fixed?
>
> Interpreting your headers according to the published FITS-WCS paper II
> should result in circular grid lines, not a rectangular grid. For the
> reasons, see section 2.8 of FITS-WCS paper II. You need to decide if you
> are going to stick with the old AIPS conventions or go with the new
> FITS-WCS conventions - in the case of a CAR projection they give very
> different results unless you have CRVAL set to (0,0).
>
> If you decide to stick with the AIPS conventions (which GAIA assumes,
> although I really think this should be a user-selectable option), I'm
> not sure what sort of plot you would expect in the case of your data.
> Would you expect dec values greater than 90 to be displayed in the top
> half of the plot?
>
>
> > Also, the projection information is not reported in the
> > title (it's written as <null>) for all of the test images.
>
> This I think must be a bug in the version of AST used by GAIA. I have no
> record of consciously fixing this bug, but CONVERT built with an earlier
> release of AST V3 shows the bug whereas CONVERT built with my current
> AST system does not.
>
>
> > Jamie reports that the MER projection seesm to fail completely. FV
> > displays it fine but it seems to display as a SIN projection in GAIA. Is
> > AST meant to support MER?
>
> I think this is the same point as for CAR. If you are expecting to see a
> rectangular grid, you wont. If you use MER, CAR, etc, the thing which is
> rectangular is *native spherical* coords (that is, a spherical
> coordinate system which has origin (0,0) at the reference point spcified
> by CRVAL). Final celestial coords (RA,Dec) will only be rectangular if
> they coincide with native spherical (i.e. if CRVAL=(0,0)). Otherwise, the
> (RA,Dec) gridlines will reflect the spherical rotation which occurs from
> native spherical to celestial.
>
> David
>
> ----------------------------------------------------------------------
> Dr David S. Berry    ([log in to unmask])
>
> STARLINK project                 |      Centre for Astrophysics
> (http://www.starlink.ac.uk/)     |      University of Central Lancashire
> Rutherford Appleton Laboratory   |      PRESTON
> DIDCOT                           |      United Kingdom
> United Kingdom                   |      PR1 2HE
> OX11 0QX
>

--
Tim Jenness
JAC software
http://www.jach.hawaii.edu/~timj