Notes

1In the original publication (Wu et al., 2006BB107), the factor in the second of the three options in equation (33) is incorrectly defined as [\textstyle{{\sum_{\rm overlap}|F_j^{o}|} / {\sum_{\rm overlap} |F_j^{\rm old}|}}]. This is a typo in the publication, and the form presented here in equation (33)[link] is the correct form used in the algorithm (Wu, 2012BB122).

2In the paper, the reported optimal ked is 1.3. This value is found if [\sigma(\rho)] is calculated only once at the begining of the iteration and is assumed to be constant during the iteration. This method was used in the version of Superflip used in the tests, and is valid for the standard CFA and for the magnitude projector (6)[link]. If, however, the Fourier magnitudes change during the iteration [as with equation (5)[link] or with the AAR scheme], [\sigma(\rho)] should be updated every cycle. With this approach it was found that optimal ked is actually 1.2 instead of 1.3 (Dumas, 2011BB121).