
On Tue 2005/09/06 11:06:30 MST, Tom Jarrett wrote
in a message to: fitsbits@xxxxxxxxxxxxxxxxx
and copied to:
>is correct. I would greatly appreciate it if someone out there has a look
>at one of our cubes, checking out the FITS header and the
>CTYPE3 = 'WAVETAB' /Wavelength
>definition. Please let me know if you find a problem or potential
>problem with what we have done. Here is the sample, it is a spectral map
It's basically correct but there are a few things you need to fix:
1) According to Paper I, if any CDi_ja cards are given then those not
given all default to zero. Therefore CD3_3 == 0 and your matrix is
singular.
2) In general, the coordinate array is of dimensionality M+1 with axis
lengths (M, K1, K2,...) which are to be recorded in a TDIMn card.
Although your table is 1dimensional (M == 1), the coordinate array
is formally considered to be 2dimensional though with a degenerate
first axis, and you should write the TDIMn card.
3) CRPIX3 and CRVAL3 are both omitted from the header which is bad form.
Also, note that TAB does not allow extrapolation; although it could be
accomodated for nondegenerate 1D tables with default indexing (perhaps
it should be) it can't easily be generalized to higher dimensions or
nondefault indexing.
Hence, your TAB only defines wavelengths for pixel coords between 1.0
and NAXIS3. If you want to compute wavelengths at pixel coordinate 0.5
or NAXIS3+0.5 you'll need to extend the coordinate array slightly and
adjust CRPIX3 and/or CRVAL3 accordingly.
Mark Calabretta
ATNF

