question for gradient scaling and adjoint source in inverse design
-
-
Hi! I'm new to inverse design and I have some questions when using Photonic Inverse Design Python API, I would greatly appreciate it if I could receive your answers and clarification!
First question is how to set up an adjoint light source in FDTD? According to the literature cited on the official website(Photonic Inverse Design Overview - Python API)，the adjoint method needs to use dipole source at the output, but all the inverse design examples that Lumerical provides just use ordinary light source(totally the same as the input light source)as the adjoint source in the adjoint simulation. So I have a specific objective function f(E,H) , and then what should I do to set up an adjoint light source for simulation?
Second question is how to scale the gradient when using L-BFGS-B as optimization algorithm? there is a parameter I need to set up before running the optimization in lumopt:
I want to know what really happened in the code, and I found this fuction to get a scaling factor for fom and gradient:
after getting the fom_scaling_factor, it just multiply this factor with fom and gradient in each iteration, so I have question that:
1.why the max_change is calculated by subtracting gradient from startpoint? why not the biggest difference between startpoint and 0/1?,like if the startpoint is 0.4, then the max_change = 1-0.4 = 0.6 ?
2.since this scaling factor is used for rescale the gradient, why it can force the optimization parameters to change at least this much, what's the relation between these two parameters?
Looking forward to your answer, thank you!
-
-
-
- The topic ‘question for gradient scaling and adjoint source in inverse design’ is closed to new replies.
You are navigating away from the AIS Discovery experience