Validation of iDPC routine

[uellue]

In principle, iDPC is supposed to calculate the potential from a field, i.e. integrate it. By using the iDPC routine instead of a simple path integral, it is supposed to reject irregular components, such as potential differences along a closed path.

With SSB and WDD, an object can be reconstructed quantitatively, i.e. the result is very close to a bandpass-filtered version of the object.

When calculating the field from a potential and then reconstructing it with iDPC, a blurred image is obtained:

Transfer function on a larger array of random data: np.abs(np.fft.fftshift(np.fft.fft2(rec_potential)/np.fft.fft2(potential)))

It seems like the weight of the individual spatial frequencies is off?

So, first, is the assumption right that potential -> field -> potential is supposed to work correctly?

In a simulated 4D STEM setting where WDD can reconstruct the object very faithfully, iDPC simply fails:

Now, is the simulation and WDD routine in https://github.com/LiberTEM/LiberTEM-iCoM/blob/9e47352a1ada0a096abcb2fa7e6508179a4c7d8f/examples/reference-validation.ipynb wrong, is this breaking assumptions of iDPC, is the iDPC function not implemented correctly, or is iDPC simply not quantitative?