spectrum calculation issue

[jasercion]

Jean has reported an issue with the calculation of a SED point when attempting to generate a spectrum of the Crab. Details can be found on confluence here: https://confluence.slac.stanford.edu/spaces/SCIGRPS/pages/587614420/gtlike+on+uw1617#gtlikeonuw1617-update0126Update3February2026

[jasercion]

It seems to have something to do with the pixel size being larger than the source and the extended source calculation ends up going off the rails, but I'm not sure what exactly is happening beyond the basics. Wasn't sure if it seemed to be recalculating the unbinned SED point using all the photons from the source (when the others are binned)?
@joeastro does anything stand out to you in particular as the likely cause/point of failure?

[jasercion]

Assigned medium priority - according to Jean it's a corner case.

[jballet]

Here is a more detailed description of the issue.
It arises in unbinned mode when fitting a small extended source, like the Crab Nebula.
It looks like the numeric integration required to compute Npred from flux for extended sources is carried out using the pixels of the exposure map for unbinned mode.
The Crab nebula is approximately a Gaussian with sigma = 0.02°, but I used 0.1° pixels in the exposure map, so the fit result was completely wrong (a factor 10 too high flux).
It takes a very long time to compute the exposure map for unbinned mode, and for all other purposes (computing Npred for the diffuse emission, mostly) 0.1° is good enough.
I rebuilt the exposure map for unbinned mode with the pixel size set to 0.02°, and the fit result went back to normal.
I propose two solutions:

1. In the short term, make it clear in the documentation of gtlike in unbinned mode that the pixel size of the exposure map must be small enough to sample the smallest extended sources present in the model.
2. In the longer term, since the exposure map is very nearly flat except at the edges, change the code to extract the exposure from interpolation in the exposure map, and define the sampling from the extended source definition when it is smaller than that of the exposure map.

[jasercion]

Thanks Jean, appreciate the explanation - issue makes a lot more sense to me now. I'll go ahead and get in touch with some folks about the documentation update and I'll add the longer term code change to our code development planning.

[jasercion]

PR #141 has the updated warning for the gtexpmap prompt. @nmirabal will be updating the online documentation.

[jasercion]

Undoing Auto-Close