Is there NAN, or BAD_VALUE in libsharp/libpsht

[Gouedard]

Hi sharp users,
I would like to know if there is NAN or BAD_VALUE defined for ignored pixels in libsharp/libpsht like in HEALPIX (BAD_DATA= -1.637500000000E+30 / Sentinel value given to bad pixels).
If YES, is it possible to change this value without consequenses.
If No, is it easy to set up one? ( a value in maps the code will ignore?)
( I am using VTK which has a different NAN).

Best regards

Claude Gouedard

[Martin Reinecke]

Hi Claude!

I would like to know if there is NAN or BAD_VALUE defined for ignored pixels
in libsharp/libpsht like in HEALPIX (BAD_DATA= -1.637500000000E+30 / Sentinel
value given to bad pixels). If YES, is it possible to change this value
without consequenses. If No, is it easy to set up one? ( a value in maps the
code will ignore?) ( I am using VTK which has a different NAN).

There is no support for special NaN-indicating pixel values in the libraries,
because they do not have a useful meaning in an SHT context. If you have a map
that has some pixels with unknown/undefined value, this corresponds to
harmonic coefficients which are all unknown/undefined. Of course the user
can decide that undefined pixels should be treated as having the value 0; but
this decision has to be made deliberately, and on a case-to-case basis.
The best way of doing this is to change the map values accordingly before
calling the SHT.
In principle I expect that this is in close analogy to FFT algorithms; there
is no FFT out there which will give meaningful results if there are NaNs in
the input.

Cheers,
Martin

[Gouedard]

Martin,
Thanks for quick answer.
Of course there is no meaning for Nan in a chart, unless 0.
It was just for using the trick, in case ...
Yes, I have to deliberatly put zeros, padding with zeros has a sens in FFT.
Many thanks.

Claude G.

[Martin Reinecke]

Hi Claude!

Of course there is no meaning for Nan in a chart, unless 0.

The missing data need not necessarily be 0, and that is exactly the
problem. Imagine you have a map of the CMB emission, with a few pixels
that are unobserved. The values of the observed pixels will be around
2.725K. If you want to do a meaningful spherical harmonic analysis of
such a map, your best bet is to fill the missing pixels with 2.725, and
not with 0. This is why the process cannot be automated by the library.

Cheers,
Martin

[Gouedard]

You are right, of course.
We can also imagin some inpainting to replace missing zones with data all arround.
I agree.
Claude G.