Adjoint method for harmonic generation optimization

[Nikolaos-Matthaiakakis]

Hello, I was curious if tidy3D supports nonlinear optimization with the adjoint method, or should someone first optimize an equivalent linear problem and then solve the nonlinear problem with the final optimized result?

Also how flexible is tidy3D with the figure of merit? can i optimize for ellipticity or for example for achieving a quadratic phase profile over a broadband range?

Thank you

[momchil-flex]

Hello, I was curious if tidy3D supports nonlinear optimization with the adjoint method, or should someone first optimize an equivalent linear problem and then solve the nonlinear problem with the final optimized result?

Gradients for nonlinear media (i.e. media with a nonlinear_spec are not currently supported (as of v2.10.0). I think taking the second approach is in general the better approach when optimizing nonlinear devices though...

Also how flexible is tidy3D with the figure of merit? can i optimize for ellipticity or for example for achieving a quadratic phase profile over a broadband range?

When it comes to optimizing linear response, Tidy3D is very flexible! You can write very complicated objective functions that combine multiple monitors or even results from multiple simulations. You can use all sorts of numpy math as long as you use import autograd.numpy as np instead of import numpy as np. We have many examples that illustrate this.

[Nikolaos-Matthaiakakis]

I understand, thank you for the reply