Missing ifftshift in filter code?

[smartalecH]

I think we are missing an ifftshift in the adjoint filter code:

meep/python/adjoint/filters.py

Lines 99 to 106 in 6abb8b3

	(kx, ky) = x.shape
	h = _proper_pad(h, 3 * np.array([kx, ky]))
	h = h / npa.sum(h) # Normalize the kernel
	x = _edge_pad(x, ((kx, kx), (ky, ky)))
	return _centered(
	npa.real(npa.fft.ifft2(npa.fft.fft2(x) * npa.fft.fft2(h))), (kx, ky)
	)

per the discussion we had awhile back.

While writing a 1D variant of this, I noticed that the _centered function ends up grabbing the ends of the filtered array (i.e. all the padding) because we never shifted the origin... but maybe numpy's 2D fft has a different convention?

(cc @mawc2019, @mochen4)

[mawc2019]

I think we are missing an ifftshift in the adjoint filter code

I do not think fftshift or ifftshift is missing. Here the kernel is padded with zeros at the center, in contrast to the previous case (before PR #2016) where the kernel is padded with zeros at ends.

While writing a 1D variant of this, I noticed that the _centered function ends up grabbing the ends of the filtered array (i.e. all the padding) because we never shifted the origin... but maybe numpy's 2D fft has a different convention?

Are you referring to the case before or after PR #2016? After PR #2016, the 1d variant was used, but fftshift or ifftshift was not used. Before PR #2016, the 1d variant was not used, but fftshift was used.

[smartalecH]

Ah yes you're right. I forgot we padded the center, not the edges. Thanks!