Learning Forum

[Luigi Ranno]

Hello,

 

I have recently been trying to make the topology optimization codes provided for the co-optimization of an asymmetric y-splitter which should send more light to one port than the other. 

I tried using the same code provided at https://lumopt.readthedocs.io/en/latest/tutorial.html#co-optimization and making 2 figures of merit, one for the left output and one for the right output, which were "added" together using the + operator. I noticed that the simple addition of the 2 FoMs causes the simulation to converge to very wrong answers. More specifically, the algorithm would make one of the two arms receive too much light, and the other too little, but in such a way that the excess of the first arm is compensated by the lack of the second, hence cancelling each other out and resulting in an apparently good FoM. No matter the initial configuration I tried, the algorithm would always converge to some wrong splitting ratio.

Would it be possible to do something more clever than simply adding the FoMs, say adding their squared magnitude, to avoid such issues? 

Thank you,

Luigi

[Guilin Sun]

It seems the co-optimization works although the result is not expected.

I guess it is related to the figure of merit: when there are two FOMS with the same strucutre, proper design of the two is vital. It is obvious one takes more than the other. You may need to have a penalty factor if one is too large.  Please explore more as our example works well for one FOM.