[fitsbits] CDELTn

[Doug Tody]

On Thu, 20 Sep 2007, Thierry Forveille wrote:

> > Looks a sensible suggestion. Not using degenerate axes is perhaps even
> > more sensibile :-)
> >
> Not at all :-) Degenerate axes are actually a most natural (if perhaps
> initially slightly counterintuitive, for some) way of conveying position
> information along the missing axes.

This has long been a point of argument.  In one view a dataset can
be viewed as N-dimensional, where N is large and there can be many
"axes", any of which are potentially degenerate (single valued).
In the other view, one has an N-dimensional sampled dataset and any
number of other dataset attributes which are constant for the entire
dataset but which do not qualify as sampled axes.

The degenerate axis view might make sense for a very few things which
are often sampled (polarization, frequency/velocity, possibly time),
where a common model can be used whether or not a given physical
"axis" is sampled (Characterization in VO is similar).  But in the
general case, applied to any "image" attribute, this is a poor model.

	- Doug