Francois,

Well, my question about CDELTn=0 was not related to WCS -- a simple
multidimensional hypercube which has several NAXISn set to 1 -- a current
practice in FITS. And I'm surprised by Phil's remark:


Regardless of the value of NAXISn, computing the coordinate value at a
pixel would involve dividing by CDELTn. For example: CRVALn + (1 -
CRPIXn) / CDELTn. So it's not as harmless as it may seem.

... I always understood that the coordinate value is CRVALn +
(i-CRPIXn)*CDELTn which simply means that the value of the coordinate
along this axis is constant and equal to CRVALn, which seems to me
perfectly correct (a 1-point dimension is by definition a constant). The
formula is not reversible, but it does not be, as far as I know.

You're right, of course (except that CRPIXn doesn't have to be 1). I
made a dumb mistake, a sign error (in the exponent), and didn't spot it
until after sending the message. Since David Berry already pointed out
that the problem with CDELTn being zero is transforming from pixel to
world coordinate, I thought everyone already saw that I should have
multiplied by CDELTn rather than dividing by it. But I thought your
question was related to WCS because it was about CDELTn, which is a WCS
keyword.

Phil