Topics related to Lumerical and more

Implementing adjoined optimization in Lumerical

    • Ruzan


      I am thinking of implementing adjoined optimization in Lumerical, and I am trying to identify the best place to start.

      I previously used the SQP method, like in this example

       The issue with this method was that with 3 parameters to be optimized, I had to run from 40 to 60  FDTD simulations to reach convergence. I believe the adjoined method should reduce the number of required FDTD simulations.

      Would optimizing the LumOpt code be the best place to start? Would you suggest using Matlab or the Python version? I found the following presentation online:

      Would you recommend some other resources?


    • Taylor Robertson
      Ansys Employee

      Hello Ruzan,


      The adjoint method is already implemented in Lumerical. I don't think it will reduce the number of simulations, but it is a more sophisticated approach to doing gradient descent. You will still need a good starting point, some physical inution and experience with FDTD\python. We ship lumopt with the Lumerical install no need to use the github, this landing page is a good place to start.

      The examples can be found here.

      Good luck

Viewing 1 reply thread
  • You must be logged in to reply to this topic.