broadband grating adjoint optimization

[malus-dance]

Based on the photonics testbed metagrating and the adjoint tutorial mode converter, I am trying to optimize a broadband 2D grating unit cell.
After some progress with the objective function, it clearly degrades with the final sigmoid bias.
Also, it is unclear to me how to set the EigenmodeCoefficient to optimize the first diffraction order.
Grateful for suggestions from anyone more seasoned.

simulation.py
optimization.py
mode_converter.py
unmapped_design_weights_beta8.csv
unmapped_design_weights_beta64.csv
unmapped_design_weights_beta32.csv
unmapped_design_weights_beta256.csv
unmapped_design_weights_beta16.csv
unmapped_design_weights_beta128.csv
layout.pdf
job30to.txt