Momchil Minkov

Yeah this is the best way to do it. You can see from the field plot that the source is not perfectly injected into the waveguide. By normalizing using the...

Overall this looks great now (I agree with minor comments from Ian). To make it actually useful for people, we'd need to document the relevant Fields methods, and also add...

I can handle the proper documentation integration if you're not familiar with how sphinx works. Maybe just add a single notebook example and I'll take it from there. We don't...

Yeah so specifically the z-dependence of the permittivity is not captured well in the E-field computation, and so it causes a discontinuity in a place where there is no interface....

The issue comes from the fact that there are a lot of intermediate matrices being constructed while computing the final matrix for diagonalization. In inference, some of those intermediate variables...

All the examples in the notebooks use the frequency-domain solver, which means that the simulation itself is computed at a single frequency. Note that the `omega` parameter goes into the...

So let me clarify what the mode solver does - it finds the eigenmodes at a fixed frequency, where the eigenvalues are the effective indexes (i.e. k-vector, or propagation constant)...

We used `angler` for that https://github.com/fancompute/angler. `ceviche` kinda grew out of there, but is mostly focused on inverse design of linear structures. However, even in angler, we only include chi-3...

> It would also be nice to include auto-diff rules for these operations, but that could come later. Just a note on this: [here](https://github.com/fancompute/legume/blob/2c40563d13c866ba6829e25376f8c2386e9242ed/legume/primitives.py#L140) I have defined an `autograd` primitive...

@shoyer oh this is very interesting! I might consider adding something like this in my `autograd` implementation because storing all (or many of) the eigenvectors to get the exact gradient...