Signal value in evaluated griddedPSFModel is dependent upon oversampling factor

[bhilbert4]

While working with some griddedPSFModels yesterday, I noticed that the total signal in the evaluated PSF seems to be dependent upon the oversampling factor of the data within the model.

In the example I'm working with, I have a griddedPSFModel where the individual PSFs in the model each have total signal values very close to 1.0. But when I evaluate the model and specify flux=1, the resulting PSF has a total signal of about 0.039. This happens to be lower than the total signal in the individual model PSFs by a factor of 25, which is the square of the oversampling factor in the library. When I evaluate the model and set flux=25, the signal in the resulting PSF matches up well with that in the input PSFs.

I see the same behavior using a second library that is oversampled by a factor of 3. In that case the signal in the evaluated PSF is a factor of 9 lower than in the input PSFs. Is this behavior expected?

This notebook shows my example calculations: https://gist.github.com/bhilbert4/c2fbce2097fefd2cb07ec6dbc7e7d412

[larrybradley]

Currently, that is the expected behavior. Jay Anderson's oversampled ePSFs sum to oversampling**2 and the photutils ePSF machinery emulates Jay's work. So, GriddedPSFModel expects that the input oversampled PSFs are normalized to oversampling**2. This should be better documented. Also, I'm thinking about adding a normalization check that perhaps issues a warning if the PSFs aren't normalized in the way photutils expects them.

[mperrin]

FYI - WebbPSF PR https://github.com/spacetelescope/webbpsf/pull/311 updates the behavior of the create psf grid function in WebbPSF to output models that are consistent with the normalization as expected by photutils.GriddedPSFModel. (i.e. we added the normalization to oversampling**2).

I think since no change is needed in photutils, this issue can be closed, unless @larrybradley you want to add in the normalization check/warning mentioned just above.