When fwhm -> 0, the initial image does not converge toward a dirac

[FerreolS]

Two related issue:

• When the FWHM is small the initial image converge toward a 2x2 square and not a dirac
• when the FWHM is very small it raise a image = 0 bug

[bourgesl]

What is expected ? An image 3x3 with all zero and 1 in center ?

[FerreolS]

When FWHM is smaller than a 2 pixels then its center should be around (0.5,0.5)

I propose:

• to bound fwhm to 1 or 0.5 pixel
• to center the gaussian in (0.5,0.5) in this case. A good things should be that the center continuously move toward (0.5,0.5) as fwhm is smaller e.g. the center is:

$\frac{0.5}{1+FWHM}$

[FerreolS]

What is expected ? An image 3x3 with all zero and 1 in center ?

the expected behaviour is that a dirac is a single non zero pixel centered at pixel coordinate (0.5,0.5) , (0,0) being the center of the image with an even number of pixels

[FerreolS]

According to @emmt , in MiRA the reconstructed image is centered on the central pixel when the width in pixel of the (square) image is odd. When the number of pixel is even the image is centered on the pixel at the top right of the center of the fov.

[FerreolS]

so should we always center the image at pixel coordinate (0.5,05)?

[emmt]

You have the FITS keywords related to coordinates allong axes to do pretty anything you want. My understanding is that you want the centre of the FOV (0,0) exactly at the center of a pixel so that a "Dirac" can be as small as one pixel.

[FerreolS]

You have the FITS keywords related to coordinates allong axes to do pretty anything you want. My understanding is that you want the centre of the FOV (0,0) exactly at the center of a pixel so that a "Dirac" can be as small as one pixel.

This is the right way to do, for an image of $N$ pixels, put the center of the image in FITS header on the center of the pixel $(N/2+1, N/2+1) $ if $N$ is even or the pixel $\left( \frac{N+1}{2} , \frac{N+1}{2} \right) $ if $N$ is odd. As it is the convention for discrete Fourier transform.