image_convolver not preserving compact source flux

[o-smirnov]

So when I run a convolution as per #35, but without a primary beam, I expect isolated compact sources to get bigger, but to keep their peak brightness.

Which they don't -- the peaks get much brighter. How do you normalize the convolution kernel? If it's unity at peak, then I can see this happening, but I don't think this is the right way to do it...

[landmanbester]

Hmmm, good point. I think I changed the convention when I started making clean beams using the same function. I have made it the default to normalise gausskern in the spi_fitter branch. It should write out the clean beam so you can check what the kernel looked like

[o-smirnov]

OK this looks more sensible. Although now I realize there can never be an exact match due to pixelization etc. As it is, I can see individual sources getting brighter or fainter (+/-30% level). I guess the only way to truly verify that nothing is off systematically is to run a source finder and do a flux-vs-flux plot, aimfast style.

[landmanbester]

If I recall correctly, my initial test was to convolve a clean beam to a single resolution and check that there is no difference between the subbands afterwards. I got this down to machine precision but I only did it in this very idealistic case. I don't really see how it can go wron in the rest of the field unless there are aliasing problems

[o-smirnov]

Should be good then... I'm just realizing the relative futility of testing this on real images. What you describe sounds a more proper test.

[landmanbester]

Hmmm, as I say that... For some strange reason I hard coded the padding fraction to 10% in image convolver. This should be fine but just in case there is now a padding-frac option

[landmanbester]

Is this still a problem or can we close the issue?

[o-smirnov]

The obvious problem is solved I guess, and the tricky one needs better testing... so may as well close....