[daip] Problem in using Aips Task SPIXR to generate spectral index map {External}

[Subhashis Roy]

Dear Eric,

You may remember that a few months back, I interacted with you on using 
IMAGR to produce wideband imaging.
This exercise worked quite well for one set of GMRT data.
However, I encountered one issue to get Spectral index image when trying
to scale noise (FLUX) for Primary beam correction in SPIXR.

More details of the processes I followed:
>From the Clean images using IMAGR on each sub-band data having N facets, 
one uses FLATN to get an image cube for the whole Field of view. Join the 
images to form a cube with MCUBE. Then, correct for Primary beam using 
PBCOR. Run Trans, and then Spixr on Trans output to get the spectral index 
map. Finally, use the Spectral index map generated in the last step as 
IN3NAME, and set IMAGRPRM(17) (I am setting it as 2) in the next run of 
Imagr for wideband Clean.

Issue with SPIXR:
I get an error "SPIXR1: SPIXDO: DOnSPX ERROR  1"
when PBPARM is set.  It then destroys the output images.
For example PBPARM =0.1, 0 (PBPARM=0.5, 0 also fails).
If PBPARM=0, SPIXR works as expected.

For you to repeat the problem, I am providing the input to SPIXR below 
(for the link to the image file, see below)

-----------------
AIPS 1: INNAME     '25089J0853+3'          Input image name (name)
AIPS 1: INCLASS    'CUB.S'                 Input image name (class)
AIPS 1: INSEQ         1                    Input image name (seq. #)
AIPS 1: INDISK        6                    Input image disk unit #
AIPS 1: OUTNAME    '25089J0853+3'          Output image name (name)
AIPS 1: OUTSEQ        0                    Output image name (seq. #)
AIPS 1: OUTDISK       6                    Output image disk unit #.
AIPS 1: BLC        *all 0                  Bottom left corner of input
AIPS 1: TRC        *all 0                  Top right corner of input
AIPS 1: FLUX       6.764E-05               Use only data > FLUX
AIPS 1: REFREQ       -1                    Reference frequency in GHz
AIPS 1:                                    < 0 -> header, 0 -> 1 GHz
AIPS 1: OPTYPE     ' '                     'CURV' fit incl curvature
AIPS 1: PBPARM        0.1      *rest 0     Beam parameters:
AIPS 1:                                    (1) Cutoff: 0 -> no PB corr
AIPS 1:                                    (2) > 0 -> Use (3)-(7)
AIPS 1:                                    (3)-(7) Beam shape
AIPS 1: DPARM         3          -3.5      Flag if bad answer levels:
AIPS 1:               2.8      *rest 0     (1) N < (2) Sp <  (3) Sp >
AIPS 1:                                    (4) T/min < (5) T/max >
AIPS 1:                                    (6) T < (7) T > (8) c <
AIPS 1:                                    (9) c >
AIPS 1: CPARM         0           0.5      Flag if uncertainty >
AIPS 1:            *rest 0                 (1) Brightness
AIPS 1:                                    (2) Spectral index
AIPS 1:                                    (3) Curvature
AIPS 1: BADDISK    *all 0                  Disk to avoid for scratch
-----------

For your reference, I am keeping the primary beam corrected cube as a 
FITS file (bzipped) in my
Google drive with the following link:
https://drive.google.com/file/d/10Z-CkamfP3E8gwt8pjzmhP_A-05rkrYo/view?usp=sharing

The file is compressed, and you will need to run 'bunzip2' to uncompress 
it. Then, you will need to run TRANS with TRANSCOD '312' and can try SPIXR 
on the output of TRANS to repeat the problem with the above inputs.

Regards,
Subhashis

On Wed, 28 Jun 2023, Eric Greisen wrote:

>
>
>> In the previous mail, I forgot to attach the modified PBCALC.FOR, which is
>> now attached.
>>
>>
>
> Thanks, I printed the document that you sent and the font was so small
> I overlooked the minus signs.  Fixing it today.
>
> This is why I ask you to check my work!
>
> I should tell you about additional experience with the spectral index
> options.  An experienced user tried to take a model image and spectral index
> image from CASA and use them with OOSUB which uses much the same routines
> as in IMAGR.  After much fiddling I found that the CASA spectral index
> image ran from -700 to +900 including at pixels where there were weak but
> numerous Clean components.  This really made a mess of the model
> computation.  When the spectral index image was clipped to make only
> vaguely reasonable values and/or when the Clean components were limited
> to the stronger ones actually on or near the source, the model computation
> came out very much better.  Using OOSUB to compute the model as a uv data
> set and then looking at the spectra is a good way to evaluate how things
> are going.  With the bad values, the spectra did strange things changing
> in odd ways.
>
>
>
> Eric
>
>