As discussed in #5, the current metagrating design exhibits a sharp resonance in the S-polarization response, which makes the result inherently non-robust (to small perturbations in discretization etc) and hard to reproduce.
One option is to re-do the optimization for only the P polarization. However, it is possible that optimizing only for P will generate a sharp resonance in P, in which case some robustification method will be required — e.g. optimizing a minimax over a bandwidth (or some other set of perturbations, e.g. dilation/erosion).
As discussed in #5, the current metagrating design exhibits a sharp resonance in the S-polarization response, which makes the result inherently non-robust (to small perturbations in discretization etc) and hard to reproduce.
One option is to re-do the optimization for only the P polarization. However, it is possible that optimizing only for P will generate a sharp resonance in P, in which case some robustification method will be required — e.g. optimizing a minimax over a bandwidth (or some other set of perturbations, e.g. dilation/erosion).